Early prediction of the risk of postoperative recurrence can help formulate treatment strategies for patients with lung cancer; however, few medical techniques accurately predict cancer recurrence, which remains a challenge in clinical practice. In addition, no previous reports on prediction of postoperative recurrence of lung cancer by computed tomography (CT) imaging have discussed the performance when using non-contrast-enhanced CT images. The aim of this study was to propose a computer-aided diagnostic (CAD) system using an ensemble of multiple deep convolutional neural networks (DCNNs) for predicting postoperative recurrence in patients who had undergone lung adenocarcinoma surgery, from contrast-enhanced and non-contrast-enhanced CT images obtained preoperatively. This retrospective study included 190 patients who underwent surgery for primary lung adenocarcinoma. The patients included in the study underwent preoperative CT scanning with a mixture of methods: both contrast-enhanced and non-contrast-enhanced, only contrast-enhanced, and only non-contrast-enhanced. In this study, data augmentation increased the number of images to 20 per patient for contrast-enhanced and non-contrast-enhanced images. This CAD system was constructed through transfer learning with five pretrained DCNNs using an ensemble method. The average of the probabilities output by these DCNN models was used as the probability of the ensemble method. We defined recurrence as reappearance of lesion within 5 years after surgical resection. Individual DCNN had the best prediction performance with 72% sensitivity, 88% specificity, 76% accuracy, and area under the receiver operating characteristic curve (AUC) of 0.80. The ensemble method performed best, with 70% sensitivity, 90% specificity, 75% accuracy, and AUC of 0.82. The ensemble method using multiple DCNN models is effective in improving the AUC for predicting postoperative recurrence of lung cancer using preoperative CT images, regardless of whether the images were contrast-enhanced, non-contrast-enhanced, or a combination of contrast-enhanced and non-contrast-enhanced. The ensemble method achieves optimal results when combining a large number of non-contrast-enhanced CT images with a small number of contrast-enhanced CT images. This approach achieves highly accurate prediction of postoperative recurrence using preoperative non-contrast-enhanced CT images, allowing noninvasive preoperative assessment.