We show that when estimating a non-parametric regression model, the k-nearest-neighbour non-parametric estimation method has the ability to remove irrelevant variables provided one uses a product weight function with a vector of smoothing parameters, and the least-squares cross-validation method...