Multiple stepwise regression was used in the selection of mixed dentition variables capable of predicting the total breadth of the unerupted permanent canine and premolars. The material consisted of 77 children. Stone casts were made before and after eruption of the canine and premolars. At the first examination when the children were, on the average, 10 years old, intraoral roentgenograms were obtained of the canine and the premolars. To predict the total breadth of the upper canine and premolars the buccolingual breadth of the upper first permanent molar and measurements on roentgenograms of the breadths of the upper canine and premolars proved most useful. In the prediction of the total breadth of the lower canine and premolars the best results were obtained with measurement of the breadths of the teeth in the roentgenograms. The breadth of the incisors proved less useful as a predictor of the breadths of unerupted canines and premolars.