Different modeling methods are used in motor learning. The aim of the present study was to compare the effect of separate and mixed modeling on dynamic balance learning in young non-athlete women. For this purpose, 48 female students with mean age of (25.5 ± 3.25) years were randomly divided into 4 groups (12 people in each group) included: skilled model, self-modeling, mixed modeling (self-modeling + skilled model), and control group (physical training without model). In the acquisition phase, each participant made 10 30-second trials on a dynamic balance gauge. At this stage, the skilled model group, before performing the odd trials, watched the video of performing the skilled model. The self-modeling group watched the video of their previous trial performance trials (even trials) before the odd trials, and the mixed-modeling group, before the odd trials, watched a video of their previous performance and then the watched the skilled model; The control group did 10 trials of the physical training. The retention test was taken after 24 hours. The results of 10 * 4 ANOVA in the acquisition stage showed that all groups had a significant improvement in dynamic balance time (P <0.05). Also, the results of one-way ANOVA in the retention test showed that the dynamic balance in the skilled model group is significantly higher than the combined and control model (P <0.001); But there was no significant difference between skilled model and self-modeling. Therefore, it seems that skilled modeling and self-modeling methods can be used to improve dynamic balance.