Evidence status

This page exists to do one thing honestly: state what the research currently supports, what it doesn’t, and where credentialled researchers actively disagree, all without deferring to any training organisation’s own account of its own effectiveness. AMI, AMS, and every national association are treated here exactly as any other named party in the literature: no more authoritative than the evidence they can point to. This page will be revisited as new studies and critiques appear; it is a living status report, not a settled verdict.

In brief

Short answers to the questions this page is most often read for. Each one summarises a section below, where the full citation, figures and caveats are given.

Does Montessori education work?

The research gives no single answer, and this site gives no single verdict. Positive effects have been reported for a range of academic and non-academic outcomes, most often in preschool and elementary settings, but researchers disagree about whether the large pooled effects in the literature are robust. Montessori education has not been shown to be superior to every well-implemented alternative.

What did the 2023 Campbell systematic review find?

Randolph et al. (2023) pooled 32 studies that showed baseline equivalence between groups and found average effects of g = 0.24 on academic outcomes and g = 0.33 on non-academic outcomes, concluding that Montessori education has “a meaningful and positive impact on child outcomes”. One of its co-authors, Angeline Lillard, also led the 2025 trial below.

What did León et al. (2026) find?

This preregistered review identified 198 studies and meta-analysed 114. It found “nominally positive and conventionally large” pooled effects but judged them not robust, and attributed the large average effects in the literature largely to structural weaknesses of the evidence base. It does not conclude that Montessori education has no effect.

What did the 2025 lottery-based trial in PNAS find?

Among 588 children who applied for places at 24 public Montessori schools in the United States, being offered a place had statistically significant effects at the end of kindergarten on reading, executive function, short-term memory and theory of mind, but not on maths, vocabulary, working memory, social problem-solving or persistence, and there were no notable impacts at the end of either preschool year. Only 20.9% of families consented to take part, and kindergarten assessments are missing for 32% and 42% of the two groups.

What is still unknown?

How far the two reviews’ study pools overlap; whether either review’s pooled estimate holds up under the other’s bias diagnostics; and how far results carry over to adolescents, from public to private programmes (or the reverse), and to families who never apply for a place.

A 2023 synthesis: Randolph et al.

Randolph, J.J., Bryson, A., Menon, L., Henderson, D.K., Kureethara Manuel, A., Michaels, S., Rosenstein, D.L.W., McPherson, W., O’Grady, R. and Lillard, A.S., “Montessori education’s impact on academic and nonacademic outcomes: A systematic review,” Campbell Systematic Reviews 19(3):e1330 (2023), DOI: 10.1002/cl2.1330. Full author list confirmed via Crossref publisher metadata. Its searches, which ended in February 2020, found 2,012 articles; 32 studies met its inclusion criteria, which required evidence of baseline equivalence between the Montessori and comparison groups, through random assignment or statistical adjustment. Together they yielded 204 effect sizes across 132,249 data points. It found a composite effect size of g = 0.24 for academic outcomes (from 0.06 for social studies to 0.26 for general academic ability) and g = 0.33 for non-academic outcomes (from 0.23 for social skills to 0.41 for children’s inner experience of school). Effects were larger in randomised than in non-randomised studies, in preschool and elementary than in middle- and high-school settings, and in private than in public Montessori programmes. The authors rated the quality of evidence as high for general academic ability, language and mathematics, moderate for executive function and creativity, and low for science, social skills and inner experience of school. They found a shortage of small studies favouring traditional education, a possible sign of publication bias, but a sensitivity analysis indicated that their findings were robust, and they concluded that “Montessori education has a meaningful and positive impact on child outcomes, both academic and nonacademic, relative to outcomes seen when using traditional educational methods”.

Worth flagging directly: Angeline Lillard, who also led the 2025 lottery-based trial described below, is the review’s tenth co-author, so this synthesis and one of the researchers whose primary work bears on its question are not fully independent of each other. This doesn’t invalidate the review (co-authorship by a subject-matter expert is normal practice), but it is relevant context for weighing it. This observation is this site’s own; the 2026 review discussed below does not raise it in its published abstract.

A 2026 review reaching a sharply different conclusion

León, S.P., Lipnevich, A.A. and García-Garrido, M. (2026), “The Montessori Method Under Scrutiny: A Meta-Analytic and Metascientific Review for Greater Transparency in Education Research,” Review of Educational Research, published online 7 July 2026, https://doi.org/10.3102/00346543261458396. A preregistered systematic review and meta-analysis that identified 198 studies, 114 of them with enough quantitative data for meta-analysis. It found “nominally positive and conventionally large” pooled effects on learning outcomes in pretest–posttest and posttest-only designs, but judged those estimates not robust: they showed extreme between-study heterogeneity, substantial small-study effects and clear signs of publication bias, and inconsistent reporting of implementation fidelity, low transparency and limited experimental control further limited how far they generalise. The authors conclude that the large average effects reported in the literature are “largely attributable to structural weaknesses of the evidence base rather than to robust educational impacts”. It does not conclude that Montessori education has no effect. This site has not yet read the full text, so nothing here describes the review’s methods beyond its published abstract.

Set beside the 2023 syntheses, Randolph et al. (above) and a parallel meta-analysis by Demangeon et al. (2023, Contemporary Educational Psychology 73, 102182) that reported positive effects in five fields, this review reaches a sharply different conclusion about what the Montessori evidence base can support. Researchers currently disagree about how strong that evidence is, and two open questions that this site cannot yet answer bear directly on the disagreement:

How far do the studies analysed by León et al. and by Randolph et al. overlap?
Does either review’s pooled estimate hold up under the other review’s bias diagnostics?

This is presented here as exactly what it is: a live, unresolved disagreement between credentialled researchers, not a question that has been settled in either direction. Readers should not treat either review as the final word. This page will be updated as further responses, replications or corrections appear, and once this site has read the full text of the 2026 review.

A national lottery-based trial (2025)

Lillard, A.S., Loeb, D., Berg, J., Escueta, M., Manship, K., Hauser, A. and Daggett, E.D. (2025), “A national randomized controlled trial of the impact of public Montessori preschool at the end of kindergarten,” Proceedings of the National Academy of Sciences 122(43):e2506130122, published online 20 October 2025, https://doi.org/10.1073/pnas.2506130122. Full citation confirmed via Crossref publisher metadata.

The trial used admission lotteries at 24 public Montessori schools in the United States to compare 588 children who had applied for a place at age three: 242 were offered a place and 346 were not. At the end of kindergarten (ages five to six), intention-to-treat effect sizes (Hedges’ g, controlling for baseline scores and demographics) were statistically significant for reading (0.297), executive function (0.250), short-term memory (0.293) and theory of mind (0.222); effects on maths (0.172), vocabulary (0.107), working memory (0.160), social problem-solving (−0.055) and persistence (−0.167) were not statistically significant; and the authors report “no notable impacts” at the end of either preschool year (one significant difference at the end of the first year, on an unvalidated persistence task, reversed in the second). A cost analysis estimated that three years of public Montessori from age three to six cost school districts USD 13,127 less per child than traditional programmes, mainly because of higher child-to-teacher ratios in the two preschool years. These results carry caveats that limit how far they generalise: only 20.9% of families consented to take part (31% of those offered a place, 17% of the others); kindergarten assessments are missing for 32% of the children offered a place and 42% of the others; only about half of the children offered a place were still in Montessori in kindergarten, so the estimates measure the effect of being offered a place rather than of attending; and the trial’s lead author, Angeline Lillard, is a co-author of Randolph et al. (2023) and the author of a 2026 book on the method, The Montessori Difference (Oxford University Press), which this site has not read. This site has not reproduced the trial’s confidence intervals.

How the trial bears on the disagreement above is not yet settled. Randolph et al.’s searches ended in February 2020, so that review could not include it, and the 2026 review’s abstract does not say whether it does. The trial studied public programmes only, the sector in which Randolph et al. found smaller effects. A 2026 conference paper by members of the same team (Daggett, Lillard, Loeb and colleagues, American Educational Research Association, doi:10.3102/2277608) analyses achievement gaps in the same study; this site has not read it. No published comment on, or correction to, the trial had been found when this site checked in October 2026.

Structural limits that apply regardless of who is right

Independent of which meta-analysis proves more durable, several limits apply to Montessori research as a field and are unlikely to be resolved by any single new study:

Self-selection and generalisation. Families generally choose Montessori rather than being assigned to it. Admission-lottery studies randomise among families who have all applied for a place, which removes that choice from the comparison itself, but their results describe applicant families, who may differ from families who never apply. Low consent and missing follow-up data, as in the 2025 trial, can also weaken the balance a lottery creates.
Fidelity variation. “Montessori” is not a protected term (see Choosing a School), and implementation varies widely, from settings run closely to one training organisation’s standards to loosely “Montessori-inspired” ones. Pooling outcomes across that range can obscure real differences in what is actually being measured.
Age-band asymmetry. The evidence concentrates on early childhood (3–6) and elementary (6–12) settings, and Randolph et al. found larger effects there than in middle- and high-school settings. Evidence for adolescent (12–18) programmes, including Montessori’s own Erdkinder concept, is thin: few such programmes exist to study, and thin evidence is not evidence that they do not work.
Sector asymmetry. Randolph et al. found larger effects in private than in public Montessori programmes, and the 2025 trial studied public programmes only, so results from one sector should not be assumed to hold in the other.
No central registry. Even basic descriptive facts, such as how many Montessori schools exist worldwide, rest on organisations’ own reports rather than any authoritative count. The 2022 Global Montessori Census documented 15,763 schools, drawing on national and regional organisations’ broadest definitions of a Montessori programme; an earlier estimate of 22,000, made in 2006 by AMI and other organisations, had country-level documentation for only two countries (see International Comparative Data). This should temper confidence in any precise worldwide statistic, including the ones on this site.

The editorial line this site holds

This site gives Montessori education no single verdict, because the evidence differs by outcome, age, setting and study design. Positive effects have been reported for a range of academic and non-academic outcomes, most often in preschool and elementary settings; the national lottery-based trial found effects on reading, executive function, short-term memory and theory of mind at the end of kindergarten, but not on maths, vocabulary, working memory, social problem-solving or persistence, and no notable impacts at the end of either preschool year; and whether the large pooled effects in the literature are robust is disputed between the 2023 and 2026 reviews. Montessori education has not been shown to be superior to every well-implemented alternative, and claims in either direction (including from within the Montessori world) should be read with the limits above in mind.

Correction, 11 October 2026: an earlier version of this page described the 132,249 data points in Randolph et al. (2023) as participants; said that admission-lottery studies compare families who applied with families who did not, when they compare applicants who were offered a place with applicants who were not; described the 2022 Global Montessori Census as a count of verified schools set against advocacy-sector estimates of 20,000–22,000, which its published text does not support; called Randolph et al. (2023) the current best synthesis; and described Angeline Lillard’s studies as the field’s most frequently cited source of causal evidence, a claim this site had not checked. See the corrections log.

Correction, 10 October 2026: an earlier version of this page gave the wrong initials for two authors of León et al. (2026), omitted its DOI, and described it as re-examining the same studies as Randolph et al. (2023), as finding the earlier effects inflated, as criticising how transparently the earlier reviews reported their search strategies and as raising a concern about researcher independence. The published abstract supports none of these descriptions. See the corrections log.