VOR learning was measured immediately (A) and 2 hr after training (B). Climbing fibers were stimulated during the ipsiversive (red) or contraversive (blue) phase of the vestibular stimulus, to roughly mimic climbing fiber responses during visual-vestibular VOR-increase or VOR-decrease training, respectively. Climbing fibers were stimulated once (CF 1x) or three times (CF 3x) per cycle of the vestibular stimulus, unilaterally or bilaterally. Each bar represents the mean ± S.E.M change in VOR gain induced by each training paradigm relative to the pre-training baseline. Bars ‘outlined in bold’ are significantly different from zero (p<0.05); asterisks indicate significant difference from the vestibular-only control (*p<0.05; **p<0.01; ***p<0.001); # indicates the learned changes in VOR gain were different when the same climbing fiber stimulation was applied during the contraversive vs ipsiversive phase of the vestibular stimulus (red vs blue bars, p<0.05); one sample t-test, Wilcoxon signed rank test, or Mann–Whitney test (see ‘Materials and methods’). Numbers indicate the number of mice for each training condition. Climbing fiber stimulation alone (light grey) induced no learning (p=0.12, Wilcoxon signed rank test for immediately post-training; p=0.79, one sample t-test for 2 hr post-training), but had a significant effect when paired with the ipsiversive phase of the vestibular stimulus (p<0.0001, Kruskal–Wallis test for training condition; post-hoc Dunn’s multiple comparison tests vs vestibular-only, *p<0.05, **p<0.01, ***p<0.0001). Optogenetic VOR-decrease training with stimulation of the climbing fibers during the contraversive phase of the vestibular stimulus, to roughly mimic their response during VOR-decrease training (blue) did not induce an associative decrease in the VOR below the vestibular-only control (dark grey). Instead, there was a slight increase relative to vestibular-only immediately after training (F(3, 50) = 3.00, p<0.05, one-way ANOVA for training condition; *p<0.05, Dunnett’s multiple comparison test), which was not significant 2 hr after training (p=0.49, Kruskal–Wallis test).