This work explores statistical and hybrid dynamical-statistical seasonal predictability of precipitation in the contiguous United States for all seasons. Although we show that total seasonal precipitation and frequencies of less-than-extreme daily precipitation events are more predictable, we focus on frequencies of daily precipitation above the seasonal 90th percentile (P90). Frequency of such heavy daily precipitation is shown to be predictable due to ENSO as well as to non-ENSO forcing. Diagnostic analysis suggests that ENSO and decadal variability in the north Pacific, including possible trends, are among the main predictors of US hydroclimate in general and P90 in particular. Specification skill achieved by a statistical model based on contemporaneous SST forcing with and without an explicit dynamical atmosphere is compared and contrasted. Statistical models relating the SST forcing patterns directly to observed station precipitation are shown to perform consistently better in all seasons than hybrid models where the SST forcing is first translated to atmospheric circulation via three separate general circulation models (GCMs) and the dynamically computed circulation anomalies are statistically related to observed precipitation. Skill is summarized for all seasons, but in detail for January-March, when we show that predictable patterns are spatially robust regardless of the approach used. Much of this predictability is due to ENSO. However, we also find that non-ENSO-related predictability is significant, especially for the extreme southwest and that this is due mostly to non-ENSO interannual and decadal variability in the North Pacific SST forcing. Prognostic analysis is carried out with the purely statistical approach to analyze P90 predictability based on antecedent SST forcing. Skill at various lead times is investigated and it is shown that significant regional skill can be achieved at lead times of up to six months even in the absence of strong ENSO forcing.