Named entity recognition (NER) finds object names in documents or sentences and classifies them into given categories. However, recognizing entity names in natural-language sentences is challenging, as it requires an understanding of various contexts. Recently, many researches have tried to apply deep learning methods to NER and have improved performance. Particularly, the bidirectional long-short-term-memory with conditional random field (bi-LSTM-CRF), which is a recurrent neural network model, is considered the most accurate for NER, as it considers contexts of both directions and is not affected by gradient-vanishing. However, the sequential nature of bi-LSTM-CRF makes the model extremely slow in training and classifying. To overcome this issue of computational speed, we propose a syllable-based Korean NER method using a convolutional neural network with CRF (CNN-CRF). The experiment with three corpora shows that the proposed model achieves a similar level of performance with bi- LSTM-CRF (0.4% improvement) however, it is 27.5% faster than bi-LSTM-CRF.