Conventional economic power dispatch problem uses deterministic models, which are however not able to reflect some real situations in practical applications. Since certain inaccurate and uncertain factors are normally involved in system operations, stochastic models are more suited for investigating some of the power dispatch problems. In this paper, the stochastic model for combined heat and power (CHP) dispatch is first formulated, and then an improved particle swarm optimization (PSO) method is developed to deal with the economic CHP dispatch by simultaneously considering multiple conflicting objectives. Based on the proposed optimization method, the impact of different problem formulations including stochastic and deterministic models on power dispatch results is investigated and analyzed. Furthermore, a comparative study is carried out with respect to genetic algorithm (GA) in terms of solution quality and computational expense.