Quantification (meta-analysis) of the neuroscience evidence on schizophrenia shows very modest average differences between patient and control distributions across a great variety of measures and literatures. The strongest findings involve cognitive and psychophysiological measures. Several possible explanations for this situation are reviewed including technical immaturity, methodological variability, dimensional and multiple illness models and the nature of cognitive measurement. An argument is developed that biological subtypes and endophenotypes within the broad diagnostic category of schizophrenia underpin the meta-analytic evidence. Considerations in the use of this evidence to identify illness variants are described and four candidate subtypes are proposed. Schizophrenia is a disease that will resist biological definition until its variants are isolated and extracted from the generic patient population.