BACKGROUND:: Poor access to or inadequate health insurance contributes to disparities in cancer incidence and mortality. Cancer registry 'payer source' data is collected by many cancer registries in the United States and has been used to compare cancer outcomes across insurance types. OBJECTIVES:: We evaluated the validity of cancer registry data on patient Medicaid status against enrollment data from Medi-Cal, California's Medicaid program. METHODS:: Data from the statewide California Cancer Registry for persons under age 65 years diagnosed with 1) any cancer in 1998 and 1999 or 2) with invasive cervical cancer between 1996 and 1999 were obtained and linked probabilistically to Medi-Cal enrollment files. We compared registry Medicaid status, determined from payer source information, against linkage results and used crosstabulations to calculate sensitivity, specificity, and positive predictive value. These measures were compared across different hospital and patient characteristics and cancer types. RESULTS:: Cancer registry Medicaid status data had poor sensitivity (48%), good specificity (98%), and moderate positive predictive value (77%). Measures of validity did not vary substantially by cancer type, stage, patient age, sex, vital status, race/ethnicity, socioeconomic status, or diagnosing hospital size. Registry data undercounted the number of Medicaid patients by 52% and incorrectly assigned Medicaid as a payer to approximately 2% of patients. CONCLUSIONS:: As a result of the poor validity of cancer registry Medicaid status data, caution should be used when interpreting cancer outcomes by insurance type calculated from registry payer source data. Linkage of registry data to Medicaid enrollment files represents a more accurate means of identifying Medicaid insurance status. [ABSTRACT FROM AUTHOR]