Spatial and temporal variations of aseismic fault creep influence the size and timing of large earthquakes along partially coupled faults. To solve for a time-dependent model of creep on the Hayward fault, we invert 18 years of surface deformation data (1992–2010), obtained by interferometric processing of 52 and 50 SAR images acquired by the ERS1/2 and Envisat satellites, respectively, and surface creep data obtained at 19 alinement and 4 creepmeter stations. For multi-temporal analysis of the SAR data we developed a method for identifying stable pixels using wavelet multi-resolution analysis. We also implement a variety of wavelet-based filters for reducing the effects of environmental artifacts. Using a reweighted least squares approach, we inverted the interferometric data to generate a time series of surface deformation over the San Francisco Bay Area with a precision of better than a few millimeters. To jointly invert the InSAR displacement time series and the surface creep data for a time-dependent model of fault creep, we use a robust inversion approach combined with a Kalman filter. The time-dependent model constrains a zone of high slip deficit that may represent the locked rupture asperity of past and future M≈7 earthquakes. We identify several additional temporal variations in creep rate along the Hayward fault, the most important one being a zone of accelerating slip just northwest of the major locked zone. We estimate that a slip-rate deficit equivalent to Mw 6.3–6.8 has accumulated on the fault, since the last event in 1868.