It is a known fact that the level of dispersion of the load (reinforcement) in a composite and the characteristics of its spatial distribution has a close relationship to the material properties. In this sense, the characterization and control of the spatial distribution and / or spread of the load in the volume of the matrix in composite materials is an important problem that is currently under research on several fronts. This thesis promotes the use of the coefficient of variation of the average distance to nearest neighbors Cov(dmean) as a parameter characterizing the load distribution in composites. In this work, the limits of use of this parameter for the identification of random and homogeneous distributions of particles in 2D images are investigated. To conduct this study it was necessary to develop different programming codes in order to simulate images of characteristic spatial configurations of bodies (random, homogeneous distribution, presence of clusters, zones of exclusion), and characterize them. As a result of this research the area of application of the methodology that uses the Cov (dmean) to study the presence of random and homogeneous distributions in images is expanded to lower concentrations below 25 per cent (in area of bodies). Moreover a limit to the minimum number of bodies in images and local fields is suggested, below which the criteria offered by the value of Cov(dmean) is unreliable. A new computer tool is developed and presented: maps of Cov(dmean). It allows to study the presence of spatial distributions in local fields and other application presented in the thesis. Also, as part of this thesis, virtual tests are conducted using the finite element software OOF2 to determine properties of real and virtual composites with various types of spatial distributions of reinforcement, observing the influence of the load distribution on the final properties of the composite.