A seasonally varying Markov model is constructed in a multivariate EOF space of observed SST and sea level analysis for 1981--98. The hindcast skill of SST and sea level is estimated with a cross-validation scheme. For SST, the correlation skill is highest in the central-eastern equatorial Pacific at about 0.5 at 9 month lead. The Markov model has a marginal SST forecast skill at 3 month lead (>0.4) in the north-western Pacific (NWP) around 10 ¿N. For sea level, the correlation skill is highest in the central equatorial Pacific and the NWP around 10 ¿N at above 0.5 at 9 month lead. In the western Pacific, sea level anomalies are generally more predictable than SST anomalies. The real time forecast of sea level by the Markov model is available at the official web site of the Climate Prediction Center, NOAA (http://www.cpc.ncep.noaa.gov/pacdir). ¿ 2000 American Geophysical Union