We describe an adaptive procedure that approximates a function of many variables by a sum of (univariate) spline functions $s_m $ of selected linear combinations $a_m \cdot x$ of the coordinates \[ \phi (x) = \sum_{1 \leqq m \leqq M} {s_m ( a_m \cdot x)}. \] The procedure is nonlinear in that not only the spline coefficients but also the linear combinations are optimized for the particular problem. The sample need not lie on a regular grid, and the approximation is affine invariant, smooth, and lends itself to graphical interpretation. Function values, derivatives, and integrals are inexpensive to evaluate.