The comparative sensitivity of ordinal multiple regression (OMR) and least squares regression (LSR) to criterion variable deviations from interval scaling was investigated by way of computer simulation. LSR on raw scores and ranks was compared to OMR on raw scores, ranks and dominances. Simulated data sets varied on predictor variable correlations, amount of prediction error, weight distinctiveness and shape of rating-scale distribution. The results indicated that LSR on raw scores was most affected by discretization in all conditions. In contrast, the performance of LSR approximated that of OMR when the data were first transformed to ranks. The poor performance of LSR on raw scores was most pronounced when the data discretization resulted in a symmetrical distribution. Predictor variable correlations and amount of prediction error did not affect the pattern of results. Weight distinctiveness did not interact with the other factors.