Lung cancer is the leading cause of cancer related death in the world. Computed tomography (CT), which can provide detailed images of lung structure, makes it possible to detect lung cancer in its early stage. Regular clinical practice involves visual inspection of hundreds cross-sectional slices of a patient's CT scan for small pulmonary nodules that can manifest early lung cancer. However, radiologists routinely miss nodules due to fatigue and the error-prone nature of the work, which may ultimately lead to incorrect diagnostic decisions. It has been shown that detection performance can be improved significantly by employing a computer algorithm for pulmonary nodule identification. This dissertation is devoted to the topic of computer-aided detection (CAD) of pulmonary nodules from chest CT scans. The thesis includes several subtopics: system architecture, optimization and validation of the detection system. Among the major contributions to the topic are: design and development of a multiscale Laplacian of Gaussian-based candidate generation system, high specificity standard moments-based pulmonary vessel bifurcation filter, nonsolid nodule detection system, and a new detection system validation procedure that compensates for size measurement error and provides a more meaningful performance assessment for CAD systems. In addition, a large size-enriched dataset for CAD system evaluation was created to become a valuable resource for future research.