This study constructs a neural network model for predicting errors and fraud in financial statements and investigates its potential as a technique for auditors to detect financial statement errors and fraud. After constructing the neural network model using financial ratios and trend variables derived from financial statements, the model is applied to a test sample to compare and analyze how effectively the neural network model can detect fraudulent financial statements. The neural network learns patterns of fraudulent and non-fraudulent financial statements through a training sample, and uses the model recognized through learning to classify given financial statements. Fraudulent financial statements are those with a high risk of fraud that require attention in the auditor's audit risk assessment and consequently demand increased substantive audit procedures. Numerous prior studies have investigated whether analytical review procedures used in the audit planning stage can efficiently detect financial statement distortions and errors, but the research findings make it difficult to draw a general conclusion. Moreover, most prior studies individually examined analytical review procedures for specific accounts to investigate how effective they are in measuring audit risk. This study measures risk using neural network techniques that derive nonlinear models by comprehensively considering all available financial data. Therefore, this study is distinguished from prior research on analytical review procedures in that it classifies financial statements using the pattern recognition capability of neural network techniques. The results of this study indicate that the neural network model produces fewer false signals when fraud is not present, while providing superior signals indicating that additional investigation is needed when fraud is present. Auditors can improve audit efficiency by utilizing neural network techniques in the audit process. Considering that existing fraud risk measurement methods are no more effective than a coin toss, improvements in risk measurement using neural network techniques can bring considerable utility to audit practice.