OBJECTIVES: To evaluate the performance of two instrument variables, namely physician and nursing home facility preference, to examine the risk of death in dual eligible elderly nursing home residents using antipsychotic agents. METHODS: A retrospective cohort design involving dual eligible nursing home residents  65 years and above was used. An instrumental variable analysis was conducted to evaluate the risk of mortality within 180 days of antipsychotic exposure.  The physician’s preference was operationalized based on the most recent antipsychotic prescription initiated by the physician. Nursing home facility preference was defined as the most frequently initiated antipsychotic agent in the nursing home.  The performance of each instruments was evaluated based on the strength of association, covariate balance, and explanatory power in addition to endogeneity tests. The risk of death was modeled using extended Cox Proportional Hazard model based on two-stage residual inclusion method for instrumental variable analysis. RESULTS:    Physician preference [Odd Ratio (OR) 3.97] and nursing home facility preference (OR 4.54) were strongly associated with antipsychotic use.  The explanatory power in the multivariate models and covariate balance in the preference groups were similar with both instruments. Instrumental variable analysis involving physician preference, however, did not meet the criteria for endogeneity (Wu-Hausman F = 1.49, P = 0.22). Using nursing home facility preference as an instrument, the extended Cox model revealed that risk of death is greater among typical antipsychotic users in the initial 40 days [Hazard Ratio (HR) 2.76, 95% CI 2.11-3.62] but decreases after 40 days [HR 1.44, 95% CI 1.10-1.88] when compared to atypical users. CONCLUSIONS:   Nursing home facility preference appears to be valid instrument in the nursing home population. Evaluation of instruments is critical in implementing instrumental variable analysis for pharmacoepidemiology research.