Sea surface temperatures (SSTs) from ship observations are assimilated into the Mellor-Yamada level 2 one-dimensional model of the mixed layer using a nonlinear version of the Kalman filter, similar to the extended Kalman filter. The model is initialized from expendable bathythermograph (XBT) temperature profiles at a weather ship location and is driven using the ship's observed meteorological forcing. As the SSTs arrive, usually every 3 hours, they are blended with the model results to produce an analysis. At the end of 3 and 5 days, the analyses are compared with another XBT profile at the same location. The data assimilation improves on the model forecasts by giving greater accuracy in the 20 to 25 m near the surface, which is a large portion of the very shallow mixed layer in these data. No improvement was found below the mixed layer. ¿ American Geophysical Union 1988