The seasonal variation of M2 internal tides is investigated using 25 years of satellite altimeter sea surface height measurements from 1992–2017. This study employs a new five-step mapping technique that combines along-track high-pass filtering, harmonic analysis, plane wave analysis, and two-dimensional bandpass filtering. The altimeter data are divided into four seasonal subsets, from which four seasonal M2 internal tide models are constructed by the same mapping technique. Two time-mean models are constructed. The first one is the 25-year-coherent model constructed using all the data following the same procedure. The second one is the seasonal-mean model obtained from the vector mean of the four seasonal models. Two seasonally-variable models are derived. The first model is a step function of the four seasonal models (i.e., variable amplitude and phase). The second model is same as the first model but that its amplitude is the seasonal vector mean (i.e., variable phase and invariable amplitude). These eight models are inter-compared, and evaluated using nine years of CryoSat-2 data. The results show that the seasonal-mean and 25-year-coherent models are equivalent. The four seasonal models have larger uncertainties, because of their short seasonal data records. The four seasonal models reduce less variance than the 25-year-coherent model. Each of these seasonal models reduces the most variance in its respective season when they are evaluated using seasonally subsetted CryoSat-2 data. The seasonally-variable models may reduce more variance in the equatorial zone, and the second seasonally-variable model works better than the first one. Direct point-wise comparisons of these models reveal significant seasonal variation in regions including the equatorial zone (±90 degrees in phase), the Amazon River mouth, the west Pacific, and the Arabian Sea. In regions such as the Northeast Pacific, the seasonal variation is detectable though weak. In conclusion, (1) the global M2 internal tides are subjected to significant seasonal variation, (2) the seasonal variation is a function of location, and (3) a fraction of the seasonal variation can be corrected by seasonally-variable internal tide models.