The North American Land Data Assimilation System (NLDAS) is a multi-institutional and multi-model research project in an uncoupled land-only mode. To date, the NLDAS executes the four land models of Noah, VIC, Mosaic, and SAC. The NLDAS has two over-arching broad purposes. The first, which was the main focus of Phase I of NLDAS, is to provide analysis and reanalysis of land states (such as soil moisture and temperature, snowpack, vegetation state, and streamflow) and their associated land surface water and energy budgets. The realtime land analyses, together with their retrospective reanalysis counterparts, can be used for monitoring (such as drought and flood monitoring, or agricultural management) or for providing initial conditions for the land component of coupled regional weather and climate prediction models. The second and newer purpose is uncoupled ensemble prediction of land states by forcing the NLDAS land models with ensemble predictions of land surface forcing over forecast ranges from weeks to months. Phase I of NLDAS, which spanned the GCIP and GAPP programs preceding CPPA, focused mainly on the analysis mode, with some pilot work done on the prediction mode. The focus of this presentation and Phase II of NLDAS is the prediction mode, with emphasis on the seasonal range. Phase II was launched recently within the CPPA program via collaborations between NCEP/EMC, NWS/OHD, NASA/GSFC/HSB, NCEP/CPC, Princeton University, University of Washington (UW), University of Maryland, and other research institutions. Phase II will include 1) a long-term 25-30 year NLDAS retrospective analysis of all four land models using CPC precipitation analyses and all other surface forcing from NCEP's North American Regional Analysis (NARR), 2) a daily realtime update using NARR's realtime extension known as the Regional Climate Data Assimilation System (R-CDAS) and 3) a seasonal predictive component. This prediction component will utilize bias-corrected ensemble seasonal predictions of surface forcing from multiple global coupled climate models and empirical ensemble seasonal predictions derived from official CPC seasonal outlooks and the historical record. Bayes theory and other objective tools will be used to weight the different sources of predicted surface forcing. In this presentation some preliminary results will be presented. For example, as a pilot project for the upcoming 25-30 year NLDAS land reanalysis, an 11-year NLDAS reanalysis has been performed from the land surface forcing produced in Phase I (October 1996 - September 2006). Disregarding the first year as spin-up, we have conducted a 10-year land-surface water and energy balance analysis from the Noah, Mosaic and VIC models (the SAC model will be added soon). The results are employed to analyze temporal and spatial distributions for soil moisture, soil temperature, sensible and latent heat fluxes, streamflow, and skin temperature, which are compared with observations. For the seasonal prediction component, we have thus far ported to NCEP and demonstrated the VIC-based ensemble seasonal streamflow prediction system developed under CPPA-sponsorship at Princeton University for the east half of CONUS using ensemble seasonal forecasts of surface forcing from NCEP's Climate Forecast System (CFS). Presently, we are preparing to port to NCEP the methodology of the CPPA-sponsored development at University of Washington (UW), whereby the official seasonal forecasts of CPC are applied to generate additional ensemble members of seasonal predictions of land surface forcing for driving multiple land models over the west half of CONUS. Within the next year at NCEP, both the dynamical prediction approach of Princeton and the empirical prediction approach of UW will be applied across the entire CONUS domain for all four NLDAS land models (VIC, Noah, Mosaic, and SAC).