The effects of detection of scattered radiation in x-ray transmission CT are studied both theoretically and experimentally. It is shown that scatter induces nonlinear errors in the measurement of attenuation values which can lead to cupping, streaks, and CT number inaccuracies. It is shown that scatter effects predominate over beam spectrum hardening effects for large body parts, and that the artifact propensity is a direct function of the scatter-to-primary ratio. The presence of scatter induced streaks were demonstrated experimentally on both a third and a fourth-generation CT scanner using an appropriate water-equivalent phantom. A simple model with constant scatter background leads to a correction algorithm which was tested on several phantoms and one human pelvis. The algorithm worked well on the smaller phantoms but was less successful for the larger objects. Nevertheless, it still gave substantial improvement in the pelvic scan. We demonstrated that, at least in the pelvis, scatter is a more significant source of error than beam hardening and that improved scatter correction algorithms are needed. The consequences for the quantitative interpretation of CT numbers for clinical diagnoses are discussed.