This study identifies the most predictable patterns of winter temperature over the contiguous United States for subseasonal timescales, specifically weeks 1-2 and weeks 3-4. The influence of ENSO was removed to focus on genuinely subseasonal predictability. Using Canonical Correlation Analysis (CCA), coupled with recent advancements in significance testing and feature selection, correlated temperature variations were identified. For weeks 1-2, the most predictable modes were found to be shifted versions of the Pacific North American (PNA) pattern, while for weeks 3-4, the most predictable mode closely resembled the PNA itself. The second most predictable mode for weeks 3-4 is independent of standard indices of subseasonal predictability, suggesting a potential new source of predictability. Soil moisture was identified as a causal factor for this second mode. The Climate Forecast System version 2 (CFSv2) captures the leading mode for weeks 3-4, but not for weeks 1-2.