A very important issue in the nonlinear analysis of complex problems is to keep the solution error under control. This can be conveniently (and usually also most economically) accomplished by the application of the adaptive analysis. Moreover, adaptivity enables to capture modes of failure that would otherwise remain undetected. The key ingredient of the adaptive process is a reliable and accurate error estimation. The residual-based error estimator enables to capture both the material as well as geometrical nonlinearity of the underlying problem and is well suited for a wide range of engineering problems. The error estimator adopted in this study proved to be efficient in terms of the qualitative as well as quantitative error assessment, which was demonstrated on its integration into the fully automatic adaptive framework applied to the analysis of the behaviour of materials with strain softening. On the other hand it also proved to be very demanding in terms of the computational times and, specially in the case of nonlocal regularization, also in terms of memory requirements. This makes the adaptive solution quite expensive and for 3D problems almost unreachable. The remedy to this problem can be found in the inherent suitability of the employed error estimation for parallelization, which is subject of further research.