This paper presents preliminary results regarding the predictive validity of learning potential tests administered in an exploratory study in Rotterdam (Holland) concerning the referral of minority students to special education. The central question of the study was why some students of Surinamese, Turkish, and Moroccan origin were placed in schools for the educable mentally retarded (EMR) while others with comparable achievement levels were not. For reasons of comparison, a sample of Dutch students with similar (low) levels of academic achievement also participated. Two learning potential tests were constructed, the first using Raven's progressive matrices, the second consisting of number series in domino notation. The tests were then administered to a sample of elementary school students consisting of the ethnic minority and Dutch children. Although the relatively short training period (about half an hour) in each learning potential test produced a significant increase in the mean posttest scores compared to the mean pretest scores, no increase of predictive validity of the posttest scores could be shown with mathematics achievement as the criterion. Neither were the pretest scores superior in the prediction of school learning. This implies that the reason of this draw is not the criterion per se, but rather the fairly high correlation between pretests and posttests. In this respect, findings were at variance with the outcomes reported in other studies. (GC)