Splines are commonly used in nonparametric function estimation. When using splines for function approximation, the choice of knot locations is crucial. In this talk, I will present a test-based knot selection algorithm that was published in the proceedings of ISI 2019. The algorithm was originally proposed in the univariate regression framework, and was extended to logistic regression and multivariate case later. I will also talk about the extension and present some simulation results.