Key aspects of regional U.S. climate variability and change during the past century lack explanation. What, for example, are the processes and causes responsible for the observed strong seasonality in U.S. surface temperature changes as well as for the spatially inhomogeneous warming? The western U.S. has been the epicenter for warming in recent decades, particularly in spring and summer, and this has led to early snowmelt and premature maximum streamflow. At the same time, there has been a lack of warming in the central U.S., especially during summer, in spite of the warming expected in the interior continent from increasing levels of greenhouse gases in the atmosphere. Strong decadal variations of U.S. climate during the last century have confounded both the detection and the attribution of regional climate trends. Prominent among these is the relatively abrupt shift in Pacific-North American climate in the mid-1970s. Other features include the decadal swings between U.S wet regimes (1910s, 1980s-90s) and dry regimes (1930s, 1950s, 2000s). Do these events reflect internal atmospheric variability? Are they the response to decadal variations in the state of the global ocean? What has been the role of anthropogenic forcing? Identifying the factors responsible for the observed low frequency variability is a necessary step toward implementing a credible decadal prediction system and for improving climate information for decision makers. Our proposal will increase understanding of observed U.S. climate variability and change through parallel development and analysis of observational and model-generated datasets, and through systematic numerical experimentation to allow attribution of observed variability to processes and causes. In particular, we seek to identify those factors driving fluctuations in U.S. surface temperature and precipitation on the regional scale by employing a hierarchy of existing climate model simulations, as well as new experiments targeted specifically to elucidate the role of oceanic variability. We will employ a multi-model architecture and make resulting data available to the broader research community.