Black-box transformer high-frequency models can be determined via frequency sweep measurements and curve fitting based on rational functions. This procedure often leads to unacceptable model behaviors around 50/60 Hz, making it hard to properly initialize the model for high-frequency transient events. In this paper, we overcome the accuracy problem by merging the frequency sweep data with a data set calculated from a standard 50/60-Hz model. The merging is performed using low-order filters, ensuring a smooth transition between the data sets. The resulting data set is subjected to model extraction via curve fitting and passivity enforcement. The modeling procedure is demonstrated for a 300-kVA distribution transformer. The model is applied in an example of circuit-breaker transient recovery voltage (TRV) simulation, requiring good accuracy at both 50 Hz and at high frequencies. © 1986-2012 IEEE.