When a Power Amplifier (PA) device is operated at a given duty cycle (power on and off periodically), the device temperature responds accordingly with the peak and valley values occurring per cycle. Detailed transient thermal analysis is required to predict the device’s thermal characteristics at specific timeframes or at steady-state. Many thermal evaluations are conducted using the steady state condition at 100% duty cycle (power on continuously), requiring less computational time than the transient analysis, but providing conservative prediction without details of the transient response. Another shortcoming of the numerical prediction is the large amount of computational time and the inability to estimate temperatures for long cycle times. A new thermal Resistor-Capacitor (RC) network approach to predicting transient thermal responses in semiconductor packages is presented in this study. The proposed compact thermal model for a given package is a thermal RC network extracted by curve fitting the temperature response predicted by simulation to a step power input. Non-grounded Foster network is adopted for the proposed RC network, as its special structure makes it simple to change the RC topology during RC network extraction. The procedure to obtain RC values in each RC topology is iterated to get optimal RC values. The RC topology and values yielding minimum RMS error between the thermal RC network and simulation are accepted as the extracted compact thermal model for the given package. The extracted model is then applied to predict the transient temperature of a given power pulse. The thermal RC networks in both model extraction and subsequent prediction are expressed in Laplace domain first, and then inverted to the time domain. This ensures two advantages: (1) curve fitting during model extraction is simplified and accelerated; (2) the extracted model can predict the temperature responses to essentially all power pulses in practice. The proposed approach is validated on a PA module. The results show that the approach works accurately in the case of single heat source in the module. The approach combined with method of superposition can accurately predict temperature responses in the cases of multiple heat sources as well. To address further challenges of self and interactive heating in multiple heat sources, a direct fit method is also proposed. Validation results show that it is an effective alternative to predict transient temperatures of packages in specific situations.