The paper is aimed at presenting a study of the main limitations and problems influencing the robustness of diagnostic algorithms used in diagnostics of complex chemical processes and to present the selected exemplary solutions of how to increase it. The five major problems were identified in the study. They are associated with: uncertainties of fault detection and reasoning, changes of the diagnosed process structure, delays of fault symptoms formation and multiple faults. A brief description and exemplary solutions allowing increase of the robustness of diagnostic algorithms were given. Proposed methods were selected keeping in mind applicability for the on-line monitoring and diagnostics of complex chemical processes.