This study evaluated the probability of generating false positives with A-B graphs. We generated 1,000 graphs consisting of three stable A-phase data points at 25% and three random B-phase data points; 1,000 graphs consisting of three stable A-phase data points at 50% and three random B-phase data points; and 1,000 graphs consisting of three random A-phase data points and three random B-phase data points. Results indicate that false positives were produced for (a) a relatively high percentage of graphs containing nonrandom data points in the A phase and (b) less than 2% of graphs containing random data points in both the A and B phases. These findings suggest that A-B designs may be a stronger clinical tool for evaluating the effects of interventions than previously recognized.