In this study, a gear rattle prediction model that incorporates manufacturing errors such as eccentricities and tooth indexing errors is proposed. Sets of rattle experiments are performed by using gears of known eccentricities and tooth indexing errors. The model is validated through comparison to these measurements. Experimental and theoretical results show that the gear motions are likely to increase their rattle severity in the presence of such errors. Specifically, tooth indexing errors cause rattling motions to become more prominent and often non-periodic with higher rattle severity indices as use of gears having smaller tooth indexing errors is advisable for minimizing rattle. Eccentricities are found to result in cyclic variations of the backlash magnitude to cause long-period rattling motions with non-integer average number of coast-side impacts per gear rotation. Further, phasing relationships defining the relative orientations of eccentricity vectors with respect to each other and with the torque fluctuations are shown to be influential in determining whether rattle motions will occur. It is demonstrated that the eccentricities of gears can be clocked at a certain angle with respect to each other and torque fluctuations to reduce rattling. A new no-rattle criterion is proposed at the end that includes the eccentricity magnitudes and these key phasing relationships.