For motion control, learning feedforward controllers (LFFCs) should be applied when accurate process modelling is difficult. When controlling such processes with LFFCs in the form of multidimensional B-spline networks, large network sizes and a poor generalising ability may result, known as the curse of dimensionality. Therefore, a parsimonious (reduced dimensionality) LFFC is required. Empirical modelling methods are not suited to obtain parsimonious networks for highly nonlinear processes because large data sets are needed. Alternatively, (qualitative) process knowledge can be used to construct parsimonious LFF controllers. In the research reported, a parsimonious LFFC was applied to a linear motor motion system. The experiments showed fast learning, good network parsimony, and small tracking errors for a range of motions.