In our method, a Multipath Convolutional Neural Network (MP-CNN) is proposed for activity recognition using sensor data. It consists of two novel components: Dynamic Convolutional Neural Network (D-CNN) and a State Transition Probabilities CNN (S-CNN). In D-CNN, Gaussian Mixture Models (GMM) is exploited to capture the distribution of sensor data for each activity. Then, input signal and the GMMs are screened into different segments. These form multiple paths in the CNN. S-CNN uses a modified LZW algorithm to extract the transition probabilities of hidden states as discriminate features. Then, D-CNN and S-CNN are combined to build the MP-CNN. Experimental results on several activity recognition datasets demonstrate the superior performance of MP-CNN.