In order to enhance reliability of manufacturing system, this paper advances a modeling method based on problem. Firstly, this paper uses Coal - Question - Metric ( GQM) method to analyze manufacturing system structurally, constructs structural net model consists of goal - chains, question - chains and metric - chains. Secondly, this paper analyzes mapping relations among problems, error, fault and incident, and establishes causality evolution model based on problem occurrence mechanism; after that, it utilizes Bayesian networks and evidence reasoning to construct prediction model of manufacturing system reliability based on networks of metrics, questions and goals, then establishes reliability assurance model consisting of reactive, preventive, predictive and proactive problem treatment modes. In the end, this paper analyzes key technologies of manufacturing system reliability based on problem.