Short forms of the WISC-R were computed by multiple linear regression on two samples of British children aged 6 to 12 years. The predictive validity of each short form was assessed both within the sample on which it was developed and cross-validated on the other sample. The empirically determined loss of power of prediction was found to be less than expected on a priori grounds. Predictions of individual scores at the lower end of the IQ distribution were less acceptable. It is concluded that short forms are robust and of value in research and for screening purposes, but cannot be recommended for clinical purposes.