Univariate splines have been proved quite useful in practice. However, if one wants to fit a surface, or solve a partial differential equation numerically, one would naturally think of using multivariate splines. Here splines still mean piecewise polynomial functions. In this respect, a basic question is to ascertain, for a given mesh delta and a family S of splines on delta, what its optimal approximation order is. This question is challenging even for a regular triangular mesh delta, as soon as one demands that the approximating functions have a certain amount of smoothness. The report records a step toward answering the above question. (Author).