The Local Ensemble Transform Kalman Filter (LETKF) is applied to the parallelized version of the Princeton Ocean Model to estimate the states of Loop Current and eddies in the Gulf of Mexico from April/20 to July/21, 2010 when detailed in situ current measurements were available. Tests are conducted to explore the sensitivity of the LETKF estimates to different parameters, and to systematic additions of different observational datasets which include satellite sea surface height anomaly (SSHA), satellite sea surface temperature (SST), and moored ADCP’s. The results are compared against observations to assess model skills, and also against estimates based on a simpler optimum interpolation (OI) assimilation scheme. With appropriate values of parameters and observational errors, LETKF provides improved estimates of Loop Current and eddy. In particular, the Loop Current in the late spring to summer of 2010 underwent a shedding-reattachment-shedding process. It is shown that such a nonlinear behavior is more accurately captured by LETKF, but not by OI, due to the former’s time-evolving error covariance. Finally, the accuracies of 8-week forecasts initialized from the OI and LETKF analyses and forced by reanalysis winds are compared. This period is particularly challenging to forecast because, instead of a more easily simulated westward propagation at approximately the first-mode baroclinic Rossby wave speed, the newly-shed eddy propagated very slowly, stalled, and finally decayed in the eastern Gulf. Both OI and LETKF beat persistence, but the LETKF significantly improves the eddy’s position and strength throughout the 8-week forecast.