Background: Large-scale randomised controlled trials (RCT) are relatively rare in education. The present study was an attempt to scale up previous small peer tutoring projects, while investing only modestly in continuing professional development for teachers. Purpose: A two-year RCT of peer tutoring in mathematics was undertaken in one local education authority in Scotland. The relative effectiveness of cross-age vs same-age tutoring, light vs intensive intervention, and mathematics vs mathematics + reading tutoring was investigated. Programme description: The intervention was Duolog Math, a freely available tutoring method for working with mathematical problems, involving discussion following eight structured steps and including summarising and generalising at the end. Sample: Eighty-six primary schools in one council in Scotland of average socio-economic status. Children were of mixed ability and mixed gender, with few ethnic minorities or special needs in relation to background population. Macro-evaluation n = 3520. Micro-evaluation Year 1 n = 334, Year 2 n = 447 cf. comparison group n = 85. Design and methods: Almost all the primary schools in the local authority participated and were randomly allocated to condition. A macro-evaluation tested and retested using Performance Indicators in Primary Schools over a two-year period. A micro-evaluation tested and retested within each year using a criterion-referenced test of mathematical problem solving. Macro-evaluation was with multi-level modelling, micro-evaluation with descriptive statistics and effect sizes; ANCOVA and MANOVA. Results: Macro-evaluation yielded significant pre-post gains in mathematics attainment for cross-age tutoring in randomly allocated classes over both years. No other differences were significant. Micro-evaluation yielded pre-post gains for selected schools in Year 1, but no overall gains for randomly allocated schools in Year 2 compared with comparisons. However, Year 2 girls, mathematics-only groups and less able students did gain significantly. Conclusions: Duolog Math tutoring does lead to gains in mathematics attainment compared with students not participating, but only for cross-age tutoring and girls, mathematics-only groups and less able mathematicians. Effectiveness varies according to type of tutoring (cross-age being better) and type of school (selected schools being better), but not intensity of tutoring. The results of the different forms of evaluation were indeed different. (Contains 3 tables.)