Over much of the summer of 2000 a high pressure system dominated the northeastern Pacific, that helped to maintain a persistent flow pattern over much of the continental US, associated with dry conditions in the west and wet conditoins over the east. In early September the high started to erode but still maintained considerable strength (1028 hPa analyzed central pressure on 00090500). During the summer, the NCEP global ensemble quite well predicted the persistent, above normal height at extended ranges (13.5 day fcst example). Then beginning at 00090300, with the longest, 15.5 day lead time, the ensemble mean indicated the possibility of the formation of a deep low pressure system in the Gulf of Alaska, with 1000 hPa closed central pressure. During the following days the 14.5 and 13.5 day lead time ensemble mean forecasts valid at the same time (00091812) became deeper and deeper with a 996 and 992 hPa closed low respectively. These forecasts represent large changes as compared to the high pressure dominated initial conditions. The pressure, for example, drops 24 hPa from 00090500 (initial analysis) to 00091700 (13.5 day lead fcst) around 45N, 140W. The ensemble mean forecasts, after a long period of around or above normal values, call for mslp values 16 hPa below climate average near the center of the predicted low (57.5N, 157.5W). It is interesting to note that the MRF control forecasts initiated at the same time (00090400 and 00090500) as the ensemble mean forecasts shown above exhibit wild fluctuations, with the earlier run predicting a deep low (968 hPa closed isobar), while the run initialized a day later predicting a 1020 hPa high in the same area in the Gulf of Alaska. The anomalously low pressure in the ensemble mean forecasts appear despite these large fluctuations. The majority of the members apparently predict the development of a very deep low pressure system while a minority of members indicate other scenarios are also possible. The spread within the ensemble reflects these forecast differences. Notwithstanding the inherent uncertainty, there is a great deal of forecast information in the ensemble forecasts - they provide a picture that, unlike that given by the control fcsts, is quite coherent with decreasing lead time. The right way of expressing such forecast information is in terms of probabilities. And this is one of those cases when relatively high probabilities can be used to predict events at long lead time. Apparently, regime changes are forecast by these ensembles. A similarly interesting forecast case has been documented from February 1999. Whether the regime change will occur as predicted by the ensemble mean this time is still a question.