The author suggests that most research on the early detection of learning disabilities is characterisized by an ineffective and a theoretical method of selecting and validating tasks. An alternative technique is proposed, based on a neurological theory of cognitive processes, whereby task analysis is a first step, with empirical analyses as subsequent stages. The possible utility of this technique in addressing current prediction research issues is noted, such as developmental differentiation and integration, and modality specialization. (Author/SBH)