This study presents a comparative study of Extreme Value Theory (EVT) concepts, applied to wildland fire data collected in Greece over a 25-year period (1983 − 2007). The dataset comprises 28,658 fire records from the Hellenic Forest Service, including details such as coordinates and size of the burned area. The primary objective is to develop a method for quan-tifying extreme values and estimating return periods and thus translating fire size into a more interpretable measure of fire significance. The study applies and evaluates both the block maxima generalized extreme value (GEV) approach and the peaks over threshold (POT) approach, using frequentist and Bayesian frameworks. The results from the GEV distribution indicate an asymptotic leveling off at a constant value, suggesting that the burned area size remains unchanged over time. This outcome is clearly unrealistic, implying that the GEV model provides reliable predictions only for time frames closely aligned with the period during which the empirical data were collected. Although conceptually different, the Bayesian POT approach also demonstrates unrealistic behavior over longer periods, with the burned area size continuously increasing along a power-law curve. In contrast, the frequentist POT model performs well with the available data, providing a more realistic and accurate representation of the increasing burned area size over time. According to the frequentist POT approach, the largest forest fire ever recorded in Greece (Dadia Forest in 2023)—covering approximately 95,000 hectares—corresponds to a return period of 2000 years. Over the past few years, many wildland fires in Greece have resulted in exceptionally large size of burned areas.