Objective: To identify which Montreal Cognitive Assessment (MoCA) items are most sensitive to mild cognitive impairment (MCI) and Alzheimer disease (AD) and investigate the utility of an abbreviated form of the MoCA. Method: 408 subjects (normal: n = 152, MCI: n = 169, AD: n = 87) were randomly divided into two groups to derive and validate an abbreviated MoCA. Using the derivation sample, analysis of individual MoCA items determined which items were best able to differentiate the three diagnostic groups. Receiver Operating Characteristic (ROC) curves were used to evaluate and compare the diagnostic accuracy of the abbreviated MoCA, standard MoCA, and the MMSE in the derivation and cross-validation samples. Results: Serial subtraction (Cramer's V = .408), delayed recall (Cramer's V = .702), and orientation items (Cramer's V = .832) were included in the abbreviated MoCA based on largest effect sizes relative to other items. The abbreviated MoCA demonstrated a higher area under the curve (AUC) than the standard MoCA when differentiating between MCI and AD (AUC = .93 vs .89), as well as controls and AD (AUC = .98 vs .97). Additionally, the abbreviated MoCA showed a higher AUC than the MMSE when differentiating between MCI and AD (AUC = .93 vs .92), MCI and controls (AUC = .80 vs .70), and controls and AD (AUC = .98 vs .96). This general pattern of findings was confirmed in the cross-validation sample. Conclusion(s): Diagnostic accuracy of the abbreviated MoCA was generally superior to the MMSE and standard MoCA. These results suggest an abbreviated MoCA form could be an effective and efficient brief tool in detecting and characterizing cognitive impairment.