Traditional nonlinear process monitoring approaches deal with all process variables directly, without carefully investigating their complex relationships. In this Article, key performance indicators (KPIs) are monitored objects. Maximal information coefficient (MIC) is used to analyze correlation between process variables and KPIs. MIC is able to find complex associations and assess their significance in the nonlinear process data set. After relevant variables are found, orthogonal signal correction (OSC) extracts the part of the KPI that has a nonlinear relationship with these variables. In other words, KPI is divided into two parts. One part is linearly related to variables, while the other part has nonlinear relation with them. In the end, two monitoring models are built for these two parts, respectively. Applications to a numerical simulation and the Tennessee Eastman process verify the effectiveness of our proposed method.