Quantum noise in cone beam CT (CBCT) imaging was studied to provide quantitative relationships among 3D cone beam image noise level and CT acquisition and reconstruction parameters, which include entrance exposure level, number of projections, and single detector size. It showed that the level of reconstructed image noise, which was caused by quantum noise in projection data, was spatially variant and related to the shape of the scan object, and that the image noise level was inversely proportional to the square root of entrance exposure level per projection, square root of number of projections, and square of detector size. Both computer simulations and real phantom studies were conducted to verify the derived quantitative relationships between image noise level and CT parameters. Shepp-logan head phantom was used in computer simulations to verify the theoretical relation between noise level and detector size, while a real cylindrical oil-uniformed phantom was studied to verify the theoretical relation between noise level and entrance exposure level. The real phantom studies were carried out on a flat panel detector (FPD)-based CBCT system available in our Lab. This work can provide a guide on how to balance various CBCT parameters to achieve satisfactory image quality with desired signal-to-noise ratio, specified spatial resolution, low contrast detectability and minimal x-ray radiation to patients.