Because analyses of reaction-time data are sensitive to aberrant observations and violations of statistical assumptions, a new approach is suggested. In this empirical approach, one applies the same criteria to the problem of selecting a statistical method as one uses to select among alternative experimental procedures. Six criteria are presented which clarify whether or not the data are orderly, or make sense. Following are five tests, based on a set of about 36,000 observations, in which ordinary least-squares multiple regression is compared with Huber's robust iteratively reweighted least-squares method. The five tests are: (1) extent of invariance across probe delays in effect of target identity; (2) extent of invariance across probe modalities in effect of target identity; (3) deviation from linearity of array-size effect; (4) extent of invariance of array-size effect across probe modalities; and (5) clarity of model comparisons. Results favor the robust method. (Author/MDE)