The present study tested the predictive validity at 3 years of age of a screening device for the early identification of later cognitive delay. The screening device, administered between 3 and 7 months of age, is based on the infant's differential fixation "to novel" over previously shown pictures. The sample was composed of 62 infants suspected to be at risk for later mental retardation. The prevalence of delayed cognitive development (IQ less than or equal to 70) at 3 years of age was 13%. Novelty preference scores correctly identified six of eight (75%) of the delayed children. The test identified 49 of 54 (91%) of the normal children. Validity for predicting cognitive delay was 55%. Validity for the prediction of normality was 96%. The screening device proved to be equally sensitive, specific, and valid when the sample was divided into infants born at term or born preterm. The results of the present study and of a previous study indicate that detection of cognitive delay based on early novelty preferences is as easily accomplished for infants who will later be mildly delayed (IQ scores 60 to 70) as it is for those who will later be severely delayed (IQ scores less than or equal to 50). Moreover, such results are in contrast to those obtained with conventional tests tapping sensorimotor development.