Stock price prediction predicts the future trend of stocks using the previous data, which has been widely focused on. Previous works aim to use either CNN or LSTM to predict the price, and few works focus on discussing the strength and weaknesses of CNN and LSTM in stock prediction tasks. In this paper, we aim to compare both CNN and LSTM on the stock price prediction problem. We first exploit the advantages and disadvantages of CNN and LSTM. Then, we propose a combined LSTM-CNN model to achieve a better performance, which avoids the layback of LSTM and increase the robustness of CNN. Our LSTM-CNN model can provide an accurate prediction and reliable attempt to combine CNN and LSTM on the stock price prediction.