This study suggests stock price direction prediction model using audit report sentimental analysis and Convolutional Neural Network(CNN) regarding public companies listed in KOSPI. It establishes sentiment lexicon by conducting a sentimental analysis regarding the audit reports, and, utilizing the sentimental words in the audit report sentiment lexicon, it sets up a prediction model of stock price direction. Sentiment words are categorized into positive and negative meaning words in the audit report, according to the audit experts' judgements. And, using these sentiment words, it compares the performance of single/ensemble classifier and CNN. As a result, the combination between CNN and sentiment words shows significantly high quality of effectiveness. This study benchmarks single/ensemble classifiers such as regression analysis, decision tree, artificial neural network, support vector machine, random forest, adaboost, and bagging. And it also calculates the model's performance by using evaluation indices such as accuracy, precision, recall, F1-score.