In this paper propose 2D Convolutional Neural Networks in Thai sign language recognition. Network was train by end-to-end for continuous gesture recognition. 2D convolutions was choose to extract features related to the gestures of Thai sign language. The design of the 2D CNN model and the generate of Thai sign language gesture data set, using 2D convolution and pooling layers, was used to train differentiation of the data. In experiment use 3 Thai sign language gestures: "Hello", "Love" and "Sick", 1000 images of each gesture, total 3000 images were selected in training and used 100 images of each gesture, total of 300 images were selected to test. The experiment shown that the designed model can greatly enhance the gesture perception, accuracy value is 0.93 and loss value is 0.27 obtained from 2D CNNs. The total learn time is 1 hr 15 min 46 sec.