We studied 1492 children in state custody over a 6-month period to investigate the relationship between children's hospital admissions and the crisis workers' clinical assessment. A 27-item standardized decision-support tool [the Childhood Severity of Psychiatric Illness (CSPI)] was used to evaluate the symptoms, risk factors, functioning, comorbidity, and system characteristics. The CSPI has been shown to have a reliability range from 0.70 to 0.80 using intraclass correlations. Logistic regression was used to calculate age-adjusted odds ratios (AOR) of hospitalization, their 95% confidence intervals, and corresponding P values. The results showed that risk factors, symptoms, functioning, comorbidities, and system characteristics were all associated with hospital admissions. Children with a recent suicide attempt, severe danger to others, or history of running away from home/treatment settings were more likely to be hospitalized (respective AOR=12.7, P<.0001; AOR=32.3, P<.0001; AOR=3.0, P=.001). In addition, hospitalization was inversely associated with caregiver knowledge of children (AOR=0.2, P=.01) and multisystem needs (AOR=0.3, P=.04). The decision to hospitalize children psychiatrically appears to be complex. As predicted, risk behaviors and severe symptoms were independent predictors of children's hospital admissions. Interestingly, the capacity of the caregiver and the children's involvement in multiple systems also predict children's hospital admissions.