Chest radiography is the first-line test for detecting lung nodules. However, these are difficult to interpret because of problems associated with low contrast, the complexity of the thoracic region, and the small dimensions that lung nodules can present. Automated computer-aided diagnosis (CAD) systems have emerged as a second opinion tool for radiologists to increase diagnostic effectiveness and reduce workload. The present study proposes the design of a CAD system for the detection of lung nodules using machine learning. First, the images were preprocessed using methods such as convolution, local normalization filtering, and homomorphic filtering. The lung region was then segmented using a thresholding method, and nodule candidates were determined using a sliding band filter, which was segmented by applying an adaptive distance-based thresholding adaptive algorithm. Seventeen features were calculated in the regions described in the database as nodules and in others that were not nodules. Subsequently, the dimension of the feature space was reduced using the Principal Component Analysis to construct a vector of input features to a machine learning algorithm that can be effective for separating the classes. Then, a Random Forest (RF) classifier was used. The algorithm was trained and validated using images from the JSRT database, showing a sensitivity of 86.58% and a specificity of 73.7%. An external test with another database was performed, getting 100% sensitivity and 80% of specificity, which showed the generalization power of the selected model.