In this paper, two novel methods of non-linear reconstruction of magnetic resonance imaging of non-Cartesian K-Space data are presented. That is non-uniform samples at space-frequency domain. In the first place, the reconstruction method by interpolation at frequency domain, Gridding, and on the other hand, the iterative reconstruction method by means of the pseudoinversion of non-uniform fast Fourier transform with the non-linear technique of the conjugate gradient. The investigation is focused in the election of efficient parameters to minimize the reconstruction mean squared error function and its subsequent characterization through Montecarlo simulation techniques. All of this, to simplify and standardize the analysis and application of these methods, which are applied over an anthropogenic image of numerical type, known as the phantom of Shepp and Logan.