In 2004, General Motors and Toyota celebrated the 20th anniversary of the formation of their 50-50 joint venture, New United Motor Manufacturing, Inc.(NUMMI). The success of this joint venture is set in sharp contrast to the pattern of failures of strategic alliances documented in empirical research. To explain this, we recast NUMMI in terms of an alliance learning model, and organize in three sections. First, we discuss the case of NUMMI derived from published sources and our own interviews with Toyota at different stages of the joint venture; second, we develop a conceptual model depicting learning dynamics and possible extensions of extant theories; and third, as a prelude to future endeavors by both firms, this paper explores the evidence regarding what was learned, the differences in benefits derived, and the probable causes of these differences based on field interviews.