Manufacturing variations in the automated fiber placement (AFP) process are one of the causes of gaps and overlaps. These manufacturing variations can be due to robot inaccuracy, tow lateral movement on the roller, tow width variation or tow compaction. An experimental setup was built to measure and investigate these various sources of manufacturing variations and their relative contributions to gap and overlap defects. This setup consisted of a commercial AFP head instrumented with additional sensors. Among all the measured sources of variations, lateral movement of the tow on the compaction roller was the biggest contributor to gaps and overlaps. The distributions of these sources of variations were fit with probability density functions. Random samples from these fits were used to simulate adjacent tows and predict the occurrence of gap and overlap defects. The distribution of predicted gaps correlated closely with the distribution of experimentally measured gaps. Thus, this approach of using statistical information about the sources of manufacturing variations to predict the frequency and magnitude of defects in a layup was validated.