Combat always involves uncertainty and uncertainty entails risk. To ensure that a combat task is prosecuted with the desired probability of success, the task commander has to devise an appropriate task force and then adjust it continuously in the course of battle. In order to do so, he has to evaluate how the probability of task success is related to the structure, capabilities and numerical strengths of combatants. For this purpose, predictive models of combat dynamics for combats in which the combatants fire asynchronously at random instants are developed from the first principles. Combats involving forces with both unlimited and limited ammunition supply are studied and modeled by stochastic Markov processes. In addition to the Markov models, another class of models first proposed by Brown was explored. The models compute directly the probability of win, in which we are primarily interested, without integrating the state probability equations. Experiments confirm that they produce exactly the same results at much lower computational cost.