Knowledge distillation has made remarkable achievements in model compression. However, most existing methods demand original training data, while real data in practice are often unavailable due to privacy, security and transmission limitation. To address this problem, we propose a conditional generative data-free knowledge distillation (CGDD) framework to train efficient portable network without any real data. In this framework, besides the knowledge extracted from teacher model, preset labels are introduced as additional auxiliary information to train the generator. The trained generator can produce meaningful samples of specified category. These samples will be employed as training data for the distillation process. In addition to using conventional distillation loss, we treat preset label as ground truth label so that the student network is directly supervised by the category of synthetic training sample. Naturally, the additional information can facilitate distillation process. Moreover, student network is forced to mimic the attention maps of teacher model and further improve its performance. We also design a novel evaluation metric called as relative accuracy to directly compare the effectiveness of different distillation methods. Trained portable network learned with the proposed data-free distillation method obtains 99.63%, 99.07% and 99.84% relative accuracy on CIFAR10, CIFAR100 and Caltech101, respectively. The superiority of proposed method is demonstrated by these experimental results.