This paper puts forward a new maximal margin classification algorithm based on general data field (MMGDF). This method transforms the linear inseparable problem into finding the nearest points in the general data field (GDF). GDF is inspired by the physical field. Different dimensions represent the different properties. Not all attributes play a decisive role in the classification process. Therefore, how to find decisive data points and attributes is a critical issue. This research builds a general data field structure in high dimension data sets. The role of data point is expanded from local to global by GDF. We calculate the weights of data points and features by the potential value in data field space. We put it into practice. Experiments show that MM-GDF method is better than the exiting common methods.