In this paper we investigate numerically the order of convergence of an isogeometric collocation method that builds upon the least-squares collocation method presented in [1] and the variational collocation method presented in [14]. The focus here is on smoothest B-splines/NURBS approximations, i.e, having global C p-1 continuity for polynomial degree p. In particular, we show that using as collocation points a suitable subset of those considered in [1] (which are related to the Galerkin superconvergence theory) it is possible to achieve optimal L2-convergence for odd degree B-splines/NURBS approximations with a pure collocation scheme, i.e., considering as many collocation points as degrees-of-freedom. The method in [1], instead, is based on a leastsquares formulation due to the fact that the set of collocation points outnumbers the degrees-of-freedom to be computed. We especially highlight that we obtain fourth-order convergence for the L2 and Linfty norm of the error when considering cubic basis functions. Further careful analysis is however needed, since the robustness of the method and its mathematical foundations are still unclear.