A generalized chance‐constrained linear decision rule (LDR) reservoir model is presented which incorporates explicit consideration of the correlation structure of the inflow process. A decision rule which includes predicted flow as a state variable (termed the predictive rule) is contrasted with the rule assuming uncorrelated monthly flows (the independent rule) and a rule which takes knowledge of the upcoming period inflow as given (the Utopian or perfect hindsight rule). Changed feasibility conditions defining minimum and maximum reliable release goals are developed and analyzed. Earlier judgements that LDR models are conservative are examined and contrasted with the improved ability to make reliable release commitments with the predictive rule. Comparison with a target hitting model over a wide range of performance reliabilities shows the new rule to provide consistently more total scheduled releas