An industrial data sample screening method based on complex network community discovery comprises the steps of firstly using target data structure initial samples as complicated network nodes, calculating distances among the nodes, comparing the distances with interceptive threshold values to obtain an adjacent matrix representing node connecting relation, then using modularity maximization as an optimization objective, performing community discovery in the complicated network represented by the adjacent matrix, obtaining sample community division of corresponding problems under different situations, finally network node 'combination degree' evaluation indexes are provided, descending sorting is conducted on nodes in the network according to a combination degree value, a sample re-structure sample set is selected from each community according to a combination degree value average, and accordingly reduction of data sample sets is achieved under the situation that useful information in an original sample set is retained. Screened data samples are adopted to perform soft measurement, prediction and case-based reasoning, the accuracy of an established model can be further improved, and guarantee is provided for implementation of data-based optimized dispatching in the industrial process.