In the last decades there has been an increase of interest in the gravimetrics geoids determination. One of the main reasons for this is its use in the determination of ortometrics altitudes from the geometric altitudes, obtained through the Global Positioning System (GPS). There are several methods for the geoid determination, such as the Stokes' integral and its variations, least square collocation, fast collocation, and others. The least square collocation allows efficient estimation of the gravitational field, under the utilization of heterogeneous data's statistic characteristics on covariance functions forms. Therefore, the statistic interpretation of the collocation is based on the conjecture that the gravity field observations are the realization of a stochastic process and that both the observations and the signals to be estimated are centered variants. The covariance function is the central element of this process. All information related to the behavior of the gravitational field, its variability, correlation distance and anisotropies, as well as the functional relation between the many elements of the field are conditioned to efficiency in its determination. This work aims the discussion of the covariances functions tracing, from data utilization with regular and irregular distribution. Data distribution has straight implications on the estimation of the covariance function parameters. Therefore it will be seen the theoretical fundamentals of the covariance function estimation methods, followed by application.