Using more than three million Landsat satellite images, this research developed the first global impervious surface area (GISA) dataset from 1972 to 2019. Based on 120,777 independent and random reference sites from 270 cities all over the world, the omission error, commission error, and F-score of GISA are 5.16%, 0.82%, and 0.954, respectively. Compared to the existing global datasets, the merits of GISA include: (1) it provided the global ISA maps before the year of1985, and showed the longest time span (1972–2019), and the highest accuracy (in terms of a large number of randomly selected and third-party validation sample sets); (2) it presented a new global ISA mapping method, including a semi-automatic global sample collection, a locally adaptive classification strategy, and a spatio-temporal post-processing procedure; and (3) it extracted ISA from the whole global land area (not from an urban mask) and hence reduced the underestimation. The GISA can contribute to further understanding on the human's utilization and reformation to nature during the past half century.