The purpose of this study was to compare 12 commonly-used nonoverlap methods with each other and with the results of visual analysis. Data were obtained from 25 studies focused on embedded instruction and schema-based instruction and included a total of 101 graphs. Treatment effect estimates using 12 nonoverlap methods were calculated for each graph by hand or using an online calculator. Five experts conducted visual analysis of each graph. Results showed that strong agreements existed between visual analysis and PND, Tau(NOVLAP) and Tau-U when raw data were analysed, and PND, PNCD and PEM-T when categorised data were analysed. Of the 12 methods investigated, PND had the highest agreement rate with visual analysis, followed by PEM-T, PAND, PNCD, IRD, NAP and Tau(NOVLAP). Overall, visual analysis appeared to be more conservative, as most nonoverlap methods overestimated the treatment effect. Additional research is needed to replicate and potentially validate the findings of this study.