Objectives. Computed tomography (CT) after iodinated contrast agent injection is highly accurate for diagnosing liver tumors but may cause renal toxicity and allergic reaction. This study aimed to evaluate the potential role of the neural network in the differential diagnosis of liver tumors in non-enhanced CT images. Methods. We studied 164 hepatic lesions including 80 malignant tumors and 84 hemangiomas. Each suspicious tumor region in the digitized CT image was manually selected. The textural information of the sub-image was extracted and then the multilayer perception (MLP) neural network classified the tumor as benign or malignant according to auto-covariance features. In the experiment, all hepatic lesions were sampled with k-fold cross-validation (k= 10) to evaluate the performance. Results. The accuracy of the proposed computer-aided diagnosis (CAD) system for classifying malignancies was 80.5%, the sensitivity was 75.0%, the specificity was 85.7%, the positive predictive value was 83.3% and the negative predictive value was 78.3%. Conclusions. This system differentiates benign from malignant liver tumors with relatively high accuracy and is therefore clinically useful in reducing the need for iodinated contrast agent injection in CT examination. Because the neural network is trainable, it could be further optimized by including a larger set of tumor images.