An empirical investigation of methodological issues associated with evaluating treatment effect in single-subject research (SSR) designs is presented. This investigation: (1) conducted a generalizability (G) study to identify the sources of systematic and random measurement error (SRME); (2) used an analytic approach based on G theory to integrate measurement errors into subject sampling errors during statistical analyses of data from a multiple-baseline design; (3) compared this new approach with a conventional t-test to determine the extent to which conventional inferential statistics were inflated, increasing the probability of Type I errors in SSR; and (4) examined discrepancies among the dependability coefficients and the statistical tests of significance. The behavior of three preschool children with handicaps and six non-handicapped peer confederates, enrolled in an integrated preschool classroom in Pittsburgh (Pennsylvania), was observed during consecutive school days over the course of 5 months. A continuous observational recording procedure was used to code interactions among each of the three target children, their peer confederates, and the teacher. Examination of the grand means revealed that 10 behavior sequences did not occur. There were more interactions between the target children and peers, fewer interactions between the children and teachers, and fewer non-social utterances. Overall, 63% of the behavior sequences occurred less than 10% of the time during which reliability observations were made. Although the comparison between statistical procedures did not confirm that the G analytic approach would identify a significant number of Type I errors, the results confirm the direction of the effect proposed by H. K. Suen et al. (1990). SSR practitioners should: adopt a G approach to study the combined influence of SRME; report variance components and minimize the emphases on reliability coefficients; and investigate further applications of the G analytic approach. A 39-item list of references and six data tables are included. (RLC)