Neuropsychological (NP) studies in schizophrenia often require data reduction to avoid statistical type I error from multiple comparisons. Typically, this involves grouping measures into domains defined by experts a priori based on delineations validated in brain-injured but not schizophrenia samples (e.g. attention, executive functioning, memory, language visuospatial, motor). Component measures are arbitrarily selected and validity may not generalize to different samples or within the same sample over time. One solution to these problems is illustrated using neurocognitive subdomains based on recent schizophrenia literature, and validated with data from a longitudinal study (156 subjects) involving repeated NP testing (baseline--within 6 months of hospital discharge--and 6 and 18 months later). A priori subdomains were grouped and submitted to principal component analysis (PCA) at each time point. Longitudinal stability of the resulting factors was tested by computing congruency coefficients. Six stable factors were extracted having good construct, divergent and predictive validity. Five neuropsychological measures frequently studied in schizophrenia were not correlated with these factors, suggesting that they should be maintained as independent neurocognitive subdomains. Distinct factors for executive functioning, verbal memory and motor functions could not be validated; this raises concerns about conclusions of previous studies regarding the pattern, severity and correlates of specific neurocognitive functions in schizophrenia.