Considering that the Support Vector Data Description (SVDD) has low efficiency in training large scale samples, we put forward a novel directional greed incremental SVDD algorithm.Firstly, a certain amount of samples are randomly selected to train the original hypersphere by incremental SVDD, and the samples meeting the hypersphere requirement are nibbled away.Secondly, we choose the sample at the farthest distance to the former hypersphere as the new incremental sample at each time, and update the hypersphere model to make incremental SVDD growth step greater than the prior one.Finally, through iterative loop, the final SVDD is estabilished for classifying whole training samples.The simulation results show that, compared with ordinary SVDD and incremental SVDD, the new algorithm has the advantages of lower time-consuming and higher model training efficiency in the premise of guaranteed classifier performance.