Aiming at insufficient investigation of deadlock-free stochastic scheduling problem in manufacturing systems, under the stochastic production environment with uncertain job arrival, processing times and product demands, the deadlock-free stochastic scheduling is studied for KMC with limited buffers. Due to the lack of ability to describe quantitative index, the concept of cost automata is given firstly. Considering the cost of part processing, inventory and backlog simultaneously, the cost objective function is constructed based on infinite horizon Markov chain with a discount factor, a stochastic dynamic programming model is obtained by the uniformization technique, and then we analyze and derive the properties of optimal objective function. To overcome the curse of dimensionality arising from combination of discrete state-space dimensions, a heuristics approximate dynamic programming algorithm based on simulation and function approximation is proposed to solve the stochastic dynamic programming model. On the basis of the above, the deadlock-free stochastic scheduling strategy is derived to ensure the efficient operation of KMC. Finally, a case is given to demonstrate and validate the effectiveness of the dead-free scheduling strategy.