A stochastic model of atmospheric surface conditions, developed from 30 years of data at Ocean Weather Station P in the northeast Pacific, is used to drive a mixed layer model of the upper mean. The spectral characteristics of anomalies in the four atmospheric variables: air and dewpoint temperature, wind speed and solar radiation, and many ocean features, including the seasonal cycle are reasonably well reproduced in a 500-year model simulation. However, the ocean model slightly underestimates the range of the mean and standard deviation of both temperature and mixed layer depth over the course of the year. The spectrum of the monthly SST anomalies from the model simulation are in close agreement with observations, especially when atmospheric forcing associated with El Niño is included. The spectral characteristics of the midlatitude SST anomalies is consistent with stochastic climate theory proposed by Frankignoul and Hasselmann (1977) for periods up to ∼6 months. Lead/lag correlations and comp... Abstract A stochastic model of atmospheric surface conditions, developed from 30 years of data at Ocean Weather Station P in the northeast Pacific, is used to drive a mixed layer model of the upper mean. The spectral characteristics of anomalies in the four atmospheric variables: air and dewpoint temperature, wind speed and solar radiation, and many ocean features, including the seasonal cycle are reasonably well reproduced in a 500-year model simulation. However, the ocean model slightly underestimates the range of the mean and standard deviation of both temperature and mixed layer depth over the course of the year. The spectrum of the monthly SST anomalies from the model simulation are in close agreement with observations, especially when atmospheric forcing associated with El Niño is included. The spectral characteristics of the midlatitude SST anomalies is consistent with stochastic climate theory proposed by Frankignoul and Hasselmann (1977) for periods up to ∼6 months. Lead/lag correlations and comp...