Long-term care is one of the major expenses faced by many older Americans. Yet, we have only limited information about the risk of needing long-term care and the expected duration of care. The expectations of needing to receive home health care, live in an assisted living facility or live in a nursing home are essential inputs into models of optimal post-retirement saving and long-term care insurance purchase. Previous research has used the Robinson (1996) transition matrix, based on National Long Term Care Survey (NLTCS) data for 1982-89. The Robinson model predicts that men and women aged 65 have a 27 and 44 percent chance, respectively, of ever needing nursing home care. Recent evidence suggests that those earlierestimates may be extremely misleading in important dimensions. Using Health and Retirement Study (HRS) data from 1992-2010, Hurd, Michaud, and Rohwedder (2013) estimate that men and women aged 50 have a 50 and 65 percent chance, respectively, of ever needing care. But,they also estimate shorter average durations of care, resulting, as we show, from a greater chance of returning to the community, conditional on admission. If nursing home care is a highprobabilitybut relatively low-cost occurrence, models that treat it as a lower-probability, highcost occurrence may overstate the value of insurance.We update and modify the Robinson model using more recent data from both the NLTCS and the HRS. We show that the low lifetime utilization rates and high conditional mean durations of stay in the Robinson model are artifacts of specific features of the statistical modelthat was fitted to the data. We also show that impairment and most use of care by age has declined and that the 2004 NLTCS and the 1996-2010 HRS yield similar cross-sectional patterns of care use. We revise and update the care transition model, and we show that use of the new transition matrix substantially reduces simulated values of willingness-to-pay in an optimal longterm care insurance model.