Cognitive deficit was conceived by Emil Kraepelin (1919) to be a core feature of a single disease, dementia praecox. In contrast, Eugen Bleuler (1923) hypothesized that schizophrenia was the ‘common final pathway’ for an etiologically mixed group of disorders. The controversy engendered by these two great clinicians survives unscathed in the present diagnostic concepts of schizophrenia. Because schizophrenia is clinically heterogeneous and genetically complex, ICD-10 and DSM-IV diagnoses may not provide the optimal phenotypes for genetic analysis. We developed a novel, endophenotype-based approach to the search for susceptibility genes by generating a composite phenotype that integrates multiple measurements of neurocognition, neurological signs and personality traits using a variant of latent class analysis known as grade of membership. The WA Family Study of Schizophrenia sample (112 families, 388 individuals) yielded two distinct neurocognitive phenotypes, each correlated with schizophrenia: one indexing pervasive neurocognitive deficit and predominantly ‘negative’ symptoms, and one neurocognitively unimpaired, with florid ‘positive’ psychotic symptoms. The quantitative neurocognitive trait was used in linkage analysis as a liability covariate, on which each individual in the sample (affected and unaffected) was scored. A 10-cM genome scan, followed by ordered sets analysis resulted in a robust linkage signal (lod score >3.4) for the neurocognitive deficit subtype on chromosome 6p25-22, whereas the neurocognitively unimpaired subtype was definitively excluded for linkage to the same region. This work supports ‘splitting’ schizophrenia into genetically distinct subtypes and shows an increase in power resulting from use of composite quantitative endophenotypes.