This paper proposes a new method for mixed gas clasifcation based on the convolutional neural network(CNN) using transfer learning. The mixed gas clasifcation is chalenging because a gas sensor aray of mixedgases is complex and high dimensional data. Moreover, it is limited to obtain enough training datasets due tohigh data colection costs. To overcome the chalenges, the proposed method maps a gas sensor aray into ananalogous-image matrix, adopts the CNN for feature extraction from images, and uses transfer learning to spedup training and improve the performance of the CNN. The proposed method is validated using public mixturegas data from the UCI Machine Learning Repository and real data examples