The problem of diagnosing the dynamics of nonlinear systems from experimental data during switching between oscillatory modes is considered. The influence of the errors of computing the correlation characteristics caused by the presence of data fragments corresponding to different dynamic modes on the interpretation of the results is investigated. It has been shown that switching between the operating modes has a much stronger effect on the diagnosis of anti-correlated dynamics.