Detecting condenser abnormalities is essential to ensure the safe and efficient operation of thermal power plants. However, most of the existing methods do not consider relationships among multiple process variables or implement process monitoring in the space of original condenser process variables. This article proposes a safe operation monitoring method for condensers based on normal operating zones (NOZs). NOZs are high-dimensional geometric spaces formed by variation ranges under normal conditions of multiple related variables whose relationships are described by condenser gray-box models. One main challenge is how to analyze the influence of model parameter uncertainties on accuracies of the condenser safe operation monitoring. This challenge is resolved by defining the false alarm rate (FAR) and missed alarm rate (MAR) under NOZ model uncertainties that are organized from multiple sets of model parameters estimated from a large amount of condenser historical data. The NOZ model is formulated by achieving an optimal trade-off between FAR and MAR. Theoretical results for the upper bounds of FAR and MAR of NOZ models caused by parameter uncertainties are developed based on the Bayesian estimation rule. Industrial case studies are provided to demonstrate the effectiveness and practicability of the proposed method.