The hypotheses are advanced that: (1) certain types of persons at risk for suicide represent identifiable clinical models, and (2) that such models provide relatively homogeneous samples from which specific high-risk indicators can be derived for clinical application to others who represent that model. Nine-hundred and eighty-six psychiatric inpatients representing the "Nice Persons" model, including 35 suicides, were randomly divided into an index and a validation set. Statistical analysis of the index set (N = 579), using a screening procedure followed by linear discriminant and linear logistic procedures, identified 11 high-risk indicators from 184 prospectively determined variables. When applied to the independent validation set (N = 407), the 11 indicators identified the suicides in that set with a sensitivity of 46% and specificity of 88%. Though not statistically significant, this level of efficiency is clinically of considerable potential value. Most importantly, the hypothesis is testable and is shown to deserve further development.