Medication records in clinical information systems (CIS) are frequently inaccurate, leading to potentially incorrect clinical decisions and preventing valid decision support interventions. It is not known what characteristics of electronic medication records are predictive of their validity. We studied a dataset of 136,351 electronic medication records of patients admitted to two academic hospitals that were individually validated by admitting providers using novel medication reconciliation software. We analyzed the relationship between characteristics of individual medication records and the probability of record validation using a multivariable linear regression model. Electronic medication records were less likely to be validated if more time had passed since their last update (14.6% for every 6 months), if they represented an antiinfective (61.6%) or a prn (50.9%) medication, or if they were in an outpatient CIS rather than on an inpatient discharge medication list (18.1%); p<0.0001 for all. Several characteristics of electronic medication records are strongly associated with their validity. These findings could be incorporated in the design of CIS software to alert providers to medication records less likely to be accurate.