OBJECTIVES: Electronic medical records (EMRs) are used increasingly for research.  Our objectives were a) to understand the utility of an EMR oncology database compared with SEER cancer registry data and Medicare and commercial claims databases and b) to identify areas for improvement in data collection, analysis, and interpretation in clinical oncology, epidemiology, and comparative effectiveness research.  METHODS: Demographic, clinical, and treatment characteristics in the four databases were compared using six tumor types:  breast, lung/bronchus, head/neck, colorectum, prostate, and NHL.  Data imputation was performed using the hot-deck method; patient characteristics were compared using Cohen’s effect size.  We described patient and clinic inclusion criteria, treatment definitions, and purposes of each database to enable comparisons.  RESULTS: Sex and 10-year age distributions for each tumor type were similar across datasets.  The EMR oncology database had a large proportion of missing data for stage (~70%) and race (~40%), which were replaced with imputed values.  There were several differences in racial composition (<15%) and ambulatory chemotherapy treatment (<30%), and modest (<10%), differences in distribution of tumor type likely due to differences in geographic distribution of included patients and clinics.  Overall, Cohen’s effects size analyses indicated small to medium differences (w<0.3) in patient characteristics across databases.  Patients in the EMR database were more likely to receive biologics and less likely to receive hormones compared to those in the reference databases, with the largest differences (<40%) observed in prostate cancer patients, who are usually seen first or primarily by urologists.  CONCLUSIONS: Several factors must be considered when using EMRs for oncology research purposes with a target of the US cancer population, particularly when evaluating treatment patterns.  Important factors include evaluation of stage, geography, race, and medical facilities’ specialization.  EMR database utility might be enhanced through imputation, addition of specific physician notes (e.g., stage) and linkage to other data sources.