ABSTRACTThis paper applies discriminant analysis to a relatively large sample of residential mortgages‐some of which have defaulted or have been foreclosed and others of which have remained untroubled at least until the date of the last observation‐in an attempt to derive a function which would allow mortgage lenders to increase their accuracy of predicting mortgage difficulty.From the data available for testing, no improvement in predictability is obtained, because it is thought that the data may already have been badly biased by the screening process of the loan officers which had already taken place and by the post‐loan origination nature of the important variables in mortgage difficulty. The latter is confirmed by discriminant models of post‐origination variables which are successful. It is concluded that changes in home value, changes in mortgagor income, mortgage assumption, and regional economic considerations are the major explanations of mortgage dif