During the July 1975-June 1976 academic year, nonphysician abstractors collected data on 15,101 ambulatory patient visits managed by 175 residents in five pediatric training programs. The visits included in this study were those on which a general data base should be collected and visits for health maintenance, head trauma, impetigo, otitis media, tonsillopharyngitis and urinary tract infection. These data provided an opportunity to investigate the number of medical record audits required to obtain a stable estimate of resident performance for a given medical condition and the relationship of performance across a number of medical conditions. The standard errors of the mean for 844 distributions of per cent compliance scores for individual residents were calculated. The standard error was less than five in 97 per cent of the instances where ten or more records had been abstracted for a given resident. Although there was some variability of resident performance from case to case, these data indicate that ten records may be a lower bound on the number of abstracts required to provide a stable estimate of performance. The correlations across conditions showed a moderate amount of homogeneity of performance and a clustering of performance for the acute care conditions included in the study. The correlations across acute care conditions were all statistically significant and ranged in magnitude from .52 to .77. Although the relationships between general data base and health maintenance and the acute conditions were statistically significant in some instances, they were all of lower magnitude and ranged from -.20 to .36. Thus, a modest number of abstracts for only a few conditions may provide a reliable estimate of resident day-to-day performance in ambulatory care clinics.