Fall detection system has a great demand for elderly people living alone. Wi-Fi CSI (Channel State Information) based fall detection method can be used to build non-intrusive and nonspace- limited fall detection systems. In the conventional work on Wi-Fi CSI based fall detection, a classification performance degradation has been observed when data in different environments is used for learning and testing data. Also, that method can not capture accurate features of motion due to the signal distortion during the noise reduction, and it can not segment signals accurately when the SNR (Signal to Noise power Ratio) is small. In this paper, we propose a spectrogram image-based fall detection using Wi-Fi CSI. Unlike the conventional method, CSI is segmented with a certain sliding time window, and then the classifier detects fall by using the spectrogram image generated from segmented CSI. We use a CNN (Convolutional Neural Network) for binary classification of the spectrogram images of the fall and non-fall motions. We carried out experiments to evaluate the classification performance of our proposed method against the conventional one by using motion data in two different rooms for learning and testing data. As a result, we confirmed that our proposed method outperformed conventional one and reached 0.90 accuracy.