Contemporary theories of the cerebellum hold that the complex spike (CS) fired by cerebellar Purkinje cells (PCs) reports the error signal essential for motor adaptation, i.e., the CS serves as a teacher reducing the performance error. This hypothesis suggests a monotonic relationship between CS modulation and performance error: the modulation of CS responses should be maximal at adaptation onset and turn back to its pre-adaptation state when the error is nulled. An alternative viewpoint based on studies of saccades suggests that the modulation of the CS discharge builds up as performance error decreases, and maximum and stable CS modulation is found after adaptation has been completed (Catz et al. 2005). We wanted to know whether this pattern can be generalized to other forms of motor adaptation. We resorted to smooth-pursuit adaptation (SPA) as an example of cerebellar-dependent adaptation. SPA is induced by increasing or decreasing target velocity during pursuit initiation that leads to a gradual increase or decrease in eye velocity. We trained 2 rhesus monkeys and recorded CS from PC in vermal lobuli VI and VII during SPA. We find that SPA is accompanied by a pattern of CS firing, which at the onset of adaptation, i.e., when the error is large, is not modulated significantly. On the other hand, when initial eye velocity is stably increased or decreased by adaptation, the probability of CS occurrence during pursuit initiation decreases or increases, respectively. Overall, our results deviate from the predictions made by the classical error-coding concept.