This paper investigates the adaptation of different performance maps of centrifugal compressors driven by dualshaft gas turbines during operation. First the estimation of compressor, gas turbine and combined efficiency are considered. Obtaining performance maps is mainly based on fitting an empirical model to the history of past data together with the understanding of how much new information is contained in newly collected data samples. This amounts to solving a leastsquares problem which is formulated as a quadratic program using various constraints. Comparing the actual efficiency to the predicted efficiency by evaluating the previously fitted model, the algorithm decides whether the actual model is accurate enough or a model update is needed. The necessity of having an online model update comes from the fact that the efficiency maps can change due to several factors such as fouling. The algorithm is tested using industrial data from a gas compression station with five gas turbine- driven compressors. The results show the need of online adaptation and that it is possible to accurately predict the different efficiencies using the presented method without excessive model updates. The access to online updated performance maps allows to understand how well the system is performing and gives the opportunity to monitor the efficiency of a specific unit. It is possible to use the adapted performance maps for load sharing optimization with timevarying optimization models.