A sample of about 1,500 convicted male felons residing in prisons in California, Michigan, and Texas completed questionnaire booklets. A retest subsample of 252 men completed the questionnaire a second time. The questionnaire reports of arrests within a defined time period and the reports of current conviction offenses were compared to information from official records. Bias scores based on average survey-record discrepancies were computed, and estimates of reliability were made, based on retest correlations and correlations between records and questionnaire responses. Findings show that on the average, prisoners do not deny arrests and convictions. Response reliability was found to be moderately high for self-reports of convictions but uncertain for reports of arrests. Further, discrepancies between survey and record values are not predicted well by ability, memory, and demographic variables, so the study did not identify the kinds of prisoners prone to lying or other response errors. When analyzing arrest data and similar questionnaire data in the future, it is recommended that analysts (1) repeat analyses in each of the three State samples separately; (2) use transformed data and reanalyze to test the sensitivity of findings to the possible existence of a questionnaire bias positively correlated with true scores; and (3) combine unreliable variables thought to measure the same thing into scales to reduce the effects of random response error variation. Future similar surveys should not only include record checks but also additional data collection to describe the record error structure. Tabular data and the questionnaire are provided, along with 43 references. (Author summary modified)