The 40-item University of Pennsylvania Smell Identification Test (UPSIT) is an effective instrument to detect olfactory dusfunction in Parkinson's disease (PD). It is not clear, however, whether tests of this length are necessary to detect such dysfunction. Several studies have suggested that detection of certain odors is selectively compromised in PD, and that a test comprised of these odors could be shorter and more specific for this purpose. Therefore, we attempted to identify a subset of UPSIT odors that distinguish PD from controls with similar or improved test characteristics compared to the full test. The discriminatory power of each odor was examined using UPSIT data from a discovery cohort of 314 PD patients and 314 matched controls and ranked using multiple methods (including odds ratios, regression coefficients and discriminant analysis). To validate optimally discriminant subsets, we calculated test characteristics using data from two independent cohorts (totaling 306 PD and 343 controls). In the discovery cohort, multiple novel 12-item subsets (and the previously described Brief Smell Identification Test-B) performed similarly or improved upon the UPSIT and were better than 12 random items. However, in validation studies from independent cohorts, multiple subsets retained test characteristics similar to the full UPSIT, but did not outperform 12 random items. Differential discriminatory power of individual items is not conserved across independent cohorts arguing against selective hyposmia in PD. However, multiple 12-item subsets performed as well as the full UPSIT. These subsets could form the basis for shorter olfactory tests in the clinical evaluation of Parkinsonism.