We propose in this paper a new strategy to estimate surface normals from sparse data for reconstruction. Our approach is based on tensorial fields morphologically adapted to infer normals forming smooth surfaces. They act as three-dimensional structuring elements for finding precise normals. Robust orientation inference is performed by an enhanced accumulation process using the tensorial fields. The surface dedicated aspects of our propositions are suitable for smooth surface inference from noisy data. We present qualitative and quantitative results to show the behavior of the original methods and ours. A comparative discussion of these results remarks the efficiency of our extensions.