The Hayward fault is considered the most likely source of one or more major earthquakes in the San Francisco Bay area in the next few decades. Historically, at least one, and probably two, major earthquakes (about M 6.8) occurred along the Hayward fault, one in 1836 and another in 1868. Little is known about the 1836 event, but the 1868 earthquake was accompanied by a surface rupture that extended as much as 41 km along the southern part of the fault. Although the amount of surface slip in 1868 is uncertain, right slip (including afterslip) reached at least several centimeters, and possibly several decimeters in places. This paper documents the spatial variation of creep rate along the Hayward fault since the 1868 earthquake. Creep (aseismic fault slip) occurs over at least 66 km and may extend over the fault's entire 82-km length, of which about 13 km lies underwater. Creep rate seems nearly constant over decades, but short-term variations occur. We derive creep rate mainly from our own systematic surveying of offset cultural features (curbs, fences, and buildings). On each feature we solve directly for accumulated creep by using multiple linear regression. Creep rate mostly falls in the range of 3.5–6.5 mm/yr; but systematic variation occurs along strike. Fault segments with distinctly higher and lower rates generally correspond to parts of the fault most salient from the overall average alinement of the fault. Most distinctive is a 4-km-long section near the south end of the fault that creeps at about 9 mm/yr. Such a high rate has occurred there at least since the 1920s and probably since the 1868 earthquake, as indicated by an offset railroad track built in 1869. We suggest that this 9 mm/yr slip rate may approach the long-term or deep slip rate that controls average recurrence interval between major earthquakes. If so, assuming an elastic rebound model, the potential for slip in large earthquakes below the surficial creeping zone is now ∼1.1 m in the southern (1868) segment of the fault and ≥ 1.4 m in the northern (1836?) segment. Subtracting surface creep rates from a long-term slip rate of 9 mm/yr gives present potential for surface slip in large earthquakes of up to 0.8 m, with an average of 0.6 m in the northern segment and 0.4 m in the southern segment. We present a simple hypothesis for rupture potential that is compatible with historic creep rate, microseismicity distribution, and geodetic data. If seismic rupture occurs on segments 41 km long by 10 km deep (7 km fully locked, 3 km creeping), today's potential for seismic moment release is 1.4 × 10¹⁹ and 1.1 × 10¹⁹ N m for both 1836? and 1868 segments, respectively, and 2.5 × 10¹⁹ N m for both segments jointly. Converting moment to magnitude gives M_L 6.8 in the northern segment, M_L 6.7 in the southern segment, and M_L 7.0 for simultaneous rupture of both.