The invention relates to a petroleum geophysical prospecting earthquake reservoir predicting method. The method comprises the following steps: determining training samples according to a known reservoir, calculating an average value and standardizing, training and constructing classifiers to obtain pile centers with different weak classifier numbers, classes to which piles belong and combined classifier coefficients, classifying standardized to-be-classified sample data, and predicting petroleum gas distribution of the reservoir according to the class to which the standardized to-be-classified sample belongs. By utilizing the method, an attribute clustering algorithm can be well applied to reservoir petroleum gas prediction.