Application of flat-panel imagers (FPIs) in cone-beam computed tomography (CBCT) offers a promising new modality for full three-dimensional (3-D) x-ray imaging. Understanding the potential performance and fundamental limitations of such technology, however, requires knowledge of the noise characteristics of the 3-D imaging system. The noise performance of a prototype flat-panel cone-beam CT (FPI-CBCT) system is investigated empirically and theoretically in terms of voxel noise, noise-power spectrum (NPS), and detective quantum efficiency (DQE). Methods for NPS analysis common in characterizing 2-D imagers are extended to the fully 3-D case, and a general framework for NPS analysis in n dimensions is presented, including the important considerations of NPS convergence and normalization within the constraints of system linearity and stationarity. Factors affecting imager noise and NPS are numerous, including exposure, number of views, image blur, additive noise, reconstruction filter, and sampling matrix. The applicability of existing theoretical descriptions of CT voxel noise is examined. A theoretical cascaded systems model that accurately predicts the 2-D noise characteristics of the FPI is extended to describe the signal and noise transfer characteristics of the fully 3-D FPI-CBCT system employing filtered back-projection. Analysis of the fully 3-D NPS reveals features of the 3-D imaging system that might otherwise be missed and shows the significant effect of 3-D noise aliasing. Furthermore, it quantifies the performance of various FPI designs in CBCT (e.g., direct and indirect detectors, and variations therein) and provides a guide for the development of high-performance FPI-CBCT systems suited to specific clinical objectives.