In this work, a three-dimensional computational model of the gerbil ear is used to investigate the generation of the Symbol and Symbol distortion product otoacoustic emissions (DPOAEs). In order to predict both the distortion and reflection sources, cochlear roughness is modeled by introducing random inhomogeneities in the outer hair cell properties. The model was used to simulate the generation of DPOAEs in response to a two-tone stimulus for various primary stimulus levels and frequency ratios. As in published experiments, the Symbol DPOAEs are mostly dominated by the distortion component while the Symbol DPOAEs are dominated by the reflection component; furthermore, the influence of the levels and frequency ratio of the primaries are consistent with measurements. Analysis of the intracochlear response shows that the distortion component has the highest magnitude at all longitudinal locations for the Symbol distortion product (DP) while the distortion component only dominates close to the DP best place in the case of the Symbol DP. Decomposition of the intracochlear DPs into forward and reverse waves demonstrates that the Symbol DP generates reverse waves for both the distortion and reflection components; however, a reverse wave is only generated for the reflection component in the case of the Symbol DP. As in experiments in the gerbil, the group delay of the reflection component of the DPOAE is between Symbol and Symbol the forward group delay, which is consistent with the propagation of DP towards the stapes as slow reverse waves.