This paper is multi-analysis approach to rock failure using metric size rock samples. The use of large-scale models permits simulation of the seismic process (including internal rupture on several scales) and utilization of a dense network for observation of the spatial variations of several physical parameters. The experiments were performed both on solid rock blocks and on concrete blocks with artificial defects, which enabled simulation of internal shear fracture. The number of various precursors appears to rise up to failure, all of them clearly manifest at the stage of a rapid drop in the applied stress (unstable deformation). The experiment suggests that rocks under strain and prior to failure must be characterized by a heterogeneous field of strains. This means that the strain is distributed mosaically, dilatancy does not generate uniformly and areas where it occurs are likely to be structurally mosaic themselves. To reinforce the prediction of micro- and macrofailure, we have realized simultaneous processing of the obtained data, using sophisticated multidimensional orthogonal functions to represent the different precursors. The possibility to identify the early stages of microfailures and to predict the macrofailure by means of statistical complex parameters derived from data on local deformations, acoustic emissions, elastic waves velocities, electric resistivity and self electric potentials is shown. Despite a considerable dissimilarity in mechanical properties of granite basalt and concrete, the complex parameter proves morphologically identical. Parameter S1 reveals exponential rise up to failure in all cases, and parameter S2 is bay-shaped in form, which makes it more promising in terms of prognosis.