Prior studies have not examined the validity of severity of illness instruments in patients at low risk for mortality. We, therefore, examined the predictive validity of a newly developed instrument, the Nursing Severity Index in 5347 adult medical and surgical patients with musculoskeletal diagnoses admitted to an academic medical center in 1985-1988. The Index is based on aggregating 34 clinical observations which were recorded by primary nurses during patient care; observations reflect biologic, functional, cognitive and psychosocial abnormalities. Other data, including patient demographic data and outcomes were obtained from hospital data bases. We found that, among all study patients, admission Nursing Severity Index scores were highly related (p < 0.001) to in-hospital death rates-which were 0, 0.4, 0.8, 2.6, 6.7 and 23.5% in six hierarchical strata defined by the Index-and to nursing home discharge rates. In multivariate analyses, adjusting for diagnosis and other important covariates, each strata was associated with a 2.5-fold increased risk of mortality and a 1.6-fold increased risk of nursing home discharge. In addition, the Nursing Severity Index was an independent predictor (p < 0.001) of hospital charges and length of stay. We conclude that the Nursing Severity Index assesses multiple dimensions of illness, can be easily recorded during routine patient care, and accurately predicts hospital outcomes in an important 'low risk' group of patients. The validity of the Nursing Severity Index in other clinical subgroups should be further studied. © 1995.