CDKN2A (INK4a/ARF) is frequently disrupted in various types of human cancer, and germline mutations of this locus can confer susceptibility to melanoma and other tumours. However, because CDKN2A encodes two distinct cell cycle inhibitory proteins, p16^INK4a and p14^ARF (p19^Arf in mice), the mechanism of tumour suppression by CDKN2A has remained controversial. Genetic disruption of Cdkn2a(p19^Arf) (hereafter Arf) alone predisposes mice to tumorigenesis, demonstrating that Arf is a tumour-suppressor gene in mice. We mutated mice specifically in Cdkn2a(p16^Ink4a) (hereafter Ink4a). Here we demonstrate that these mice, designated Ink4a*^/*, do not show a significant predisposition to spontaneous tumour formation within 17 months. Embryo fibroblasts derived from them proliferate normally, are mortal, and are not transformed by oncogenic HRAS. The very mild phenotype of the Ink4a*^/* mice implies that the very strong phenotypes of the original Ink4a/Arf^Δ2,3 mice were primarily or solely due to loss of Arf. However, Ink4a*^/Δ2,3 mice that are deficient for Ink4a and heterozygous for Arf spontaneously develop a wide spectrum of tumours, including melanoma. Treatment of these mice with the carcinogen 7,12-dimethylbenzanthracene (DMBA) results in an increased incidence of melanoma, with frequent metastases. Our results show that, in the mouse, Ink4a is a tumour-suppressor gene that, when lost, can recapitulate the tumour predisposition seen in humans.