Heterogeneous residual variance effects on genetic parameters were examined for test day milk yields of Turkish Holsteins. A third order random regression models including the fixed, random additive genetic and permanent environmental effects were used. One of these models, RV10, residual variances is assumed to be different for each test day milk yields. The RV1 model has constant residual variance for each test day. Sequential (RV2 to RV9) and non-sequential (NRV2 to NRV9) groups of residual variances were also described in the models in order to compare estimates of variance components. The univariate analysis of milk yields for each test days was performed to define variance groups. The predicted residual variances ranged from 5.62 to 11.75 and from 5.61to 11.71 for RV and NRV models, respectively. Estimates of additive genetic variances changed between 0.55-6.76 for RV and 0.08-2.46 for NRV models. Permanent environmental variances were found between 2.36 and 18.60 for RV and 6.92 to 18.85 for NRV models. Heritability estimates varied from 0.02 to 0.43 for RV and 0.01 to 0.13 for NRV models. As a result, more accurate genetic parameter estimates are achieved by controlling the residual variances. RV10 model should be preferred to define details of the milk yield residual variances for each test day. However, RV5 model has been determined that an alternative model as compared with RV10.