Demographic prediction of juvenile delinquency has been hampered by the heterogeniety of youth samples. In an attempt to correct for sampling bias in predicting juvenile delinquency, 1,689 male and female youths(aged 12 to 19, drawn from a 6 percent systematic, census tract, random sample of Pennsylvania school youths) completed the Youth Needs Survey questionnaire which included demographic variables and a self-reported delinquency scale (SRD). Analyses of the questionnaire data were performed on the complete sample and on low, medium, and high SRD subsets. The data showed that different SRD predictive patterns were produced depending on the data set considered. For the general sample, only age (older) and sex (male) were predictive of higher SRD scores. Analysis of the general sample yielded 58.5 percent, 19.3 percent, and 54.7 percent correct classification of these respective groups into low, medium, and high SRD groups from age and sex. However, of all youths classified as high delinquent, 7.12 percent were false positives. Of all youths predictively classified as non-high delinquent, 15.2 percent were false negatives. Grade level was SRD predictive within the low delinquent group, with a slight tendency for higher grade levels to be associated with higher SRD scores. In the medium delinquency group, males were associated with higher SRD scores, and within the high delinquency group, active welfare status was predictive of higher SRD scores. (Author/BL)