OBJECTIVES: Joint replacement is one of the commonest elective procedure performed in the elderly veterans.  Database studies assume the occurrence of revision to be the same side as the most recent primary, in the absence of laterality being captures in claims databases.  A recent Medicare study found that such an assumption had a significant error rate with only 71% revision on the same side as the most recent joint replacement.    METHODS: We utilized the Veterans Affairs (VA) administrative and clinical databases from fiscal years 2002 to 2010.  We defined the joint replacement cohort based on the presence of Common Procedure Terminology (CPT) codes for total knee, hip or shoulder joint arthroplasty (TKA, THA, TSA).  We used NLP ascertainment of laterality of primary and revision joint replacement   RESULTS: The cohorts consisted of 84,495 patients with 87,495 procedures.  Mean age was 63 years, 94% were male; 84% were Caucasian, 14% were African-American and 2% were other. 73,488 had operative notes available and constituted our analytic dataset. Joint laterality was established based on text data for >98% of primary and >97% of all joint replacement cohorts. We found that 57% of the primary TKA were right, 41% were left and 1% simultaneous bilateral. Similar proportions were noted for primary THA and primary TSA. For revision TKA, THA and TSA, similar proportions were noted. CONCLUSIONS: NLP can obtain the laterality of joint replacement surgery and this can assist in improving the quality of database studies of joint replacement surgery that use the VA databases.