Abstract: The sample data was based on 6 825 pairs of earlywood and latewood tracheid length samples of 9 trees from dahurian larch ( Larix gmelinii Rupr. ) plantations located in Qitaihe Forest Bureau in Heilongjiang Province, northeastern China. Six functions were selected using nonlinear regression analysis. The Richards model y = β1 1 - exp( - β2x)β3 + ε was selected to model earlywood and latewood tracheid length based on goodness-of-fit statistics. Nonlinear mixed-effects modeling approach was used to build mixed-effects models of earlywood and latewood tracheid length based on Richards model yij = (β1 + b1i )1 - exp - (β2 + b2i )tβ3 + b3i + εij . The results showed that Richards model with parameters b1i ,b2i ,b3i as random effects showed the best performance for both earlywood and latewood tracheid length. Time series correlation structures AR(1) describe error correlation within tree well. Prediction precision of earlywood and latewood tracheid length mixed models could be improved through considering both random effects and time series correlation structures