We conducted an exploratory analysis of several prospectively obtained objective measures of disease activity to derive a predictive model of hospitalization for asthma among 310 adults, ages 18 to 50 yr, with moderate to severe asthma. Baseline characteristics associated with increased risk of hospitalization in the succeeding year include (1) prior year hospitalization, (2) moderate or severe respiratory impairment, (3) a medication regimen consistent with severe asthma, (4) a history of significant systemic steroid use, (5) maximum overnight PEF variability > 40%, and (6) mean evening PEF < 60% of predicted (relative risk = 6.5, 6.9, 8.1, 3.7, 3.0, and 3.2, respectively). Recursive partitioning analysis, depicted as a "classification tree," provided a more sensitive (94%) and specific (68%) multivariate description of the data set than either logistic regression (87 and 48%, respectively) or a simple additive risk model (46 and 93%, respectively). Patients with very high (> 50%), moderately elevated (10 to 15%), and very low (< 5%) risk of hospitalization were identified on the basis of particular combinations of prior hospitalization history, level of respiratory impairment, and medication regimen. Overnight variability and mean evening PEF measured at home over a 2-wk period proved less informative for risk stratification than respiratory impairment determined once at baseline by office spirometry. The findings warrant replication and extension in other populations with the goal of developing decision rules for risk stratification and effective interventions for risk reduction.