Frequency non-stationarity is an important statistical property for beyond the fifth generation (B5G) wireless communication channels due to extremely large bandwidths used in millimeter wave (mmWave) and teraherz (THz) communications. In this paper, a general averaged power delay profile (APDP) method is first introduced as a metric to determine the stationarity regions of wireless channels in the time, frequency, and spatial domains. Then, we apply this method to determine the frequency stationarity regions (FSRs) of channel measurements in a sub-6 GHz frequency band (2–4 GHz) and two mmWave frequency bands (14–16 GHz and 28–30 GHz). Frequency non-stationary (FnS) properties are found from all of the three bands. A novel FnS channel model is then proposed to model wireless channels in all frequency bands. We first split a FnS channel into a few frequency stationary (FS) sub-band channels and model them one by one, which are finally combined as a FnS channel model considering the cluster evolution in the frequency domain as well as the frequency consistency of sub-band channels. Simulation results show that the frequency correlation functions (FCFs) of the proposed FnS channel model can well fit those of the measured FnS channels.