Intelligent diagnosis and prognosis techniques have been widely applied in modern industrial practice. However, there still exist same limitations as follows: 1) the techniques take the identical type faults with different degradation degree as different individual fault patterns for classification and identification, which is unreasonable in practical industry application; 2) the diagnosis model based on the training with specific data lacks generalization ability under varying working conditions. Aiming at above mentioned problems, a multitask feature sharing neural network is proposed and applied to the intelligent diagnosis and prognosis of bearings. Firstly, the CNN is used to construct an adaptive feature extractor, which extracts deep features from raw vibration signals. Secondly, a multitask feature sharing diagnosis model is constructed for classification and prediction, and the fault classification and fault size prediction are realized. Finally, the proposed method is verified with the benchmark bearing dataset from Case Western Reserve University (CWRU). The experiment results show that the proposed method not only can realize the task of fault type classification and fault size prediction, but also possess strong generalization ability.