Building damage assessment with remote sensing images plays an important role in providing information for disaster rescue and reconstruction. Recently, deep convolutional networks show good ability for some remote sensing applications. However, it is difficult to obtain a large number of labeled samples for training in some cases. With regard to this problem, a transfer learning method based on deep neural networks is adopted to discriminate collapsed building from intact building in remote sensing images in this paper. Samples are obtained from high resolution remote sensing images and split into training and validation datasets. Then a collapsed building discrimination model is built based on very deep convolution networks (pretrained VGGNet via ImageNet). With the new trained VGG model, test dataset including samples of collapsed buildings and intact buildings are used to classify the two types of buildings. The preliminary experiment results show that the new trained model performed well for collapsed building discrimination.