To investigate long-term contingency learning that is free from recency-based episodic retrieval, we developed a paradigm in which two non-overlapping sets of words and colors were presented in alternating blocks of the experiment (“alternating blocks paradigm”). In each block, participants had to respond to the color of words, and each of the words appeared more frequently in one of the colors than in the other colors (color-word contingency learning). Long-term contingency learning effects were assessed only on the basis of the first occurrences of each word in a block, ruling out that the word could retrieve a recent episode from the same block. Due to the rationale of the paradigm, the last occurrence of words for these trials always dated back more than a complete block (> 30 trials). In a preregistered study (n=110), we found reliable contingency learning effects for the first occurrences of words in each block, indicating long-term learning of the contingencies that is independent of episodic retrieval. Furthermore, and in line with recent findings from our lab (Rudolph & Rothermund, 2024; Rudolph et al., in press), these long-term contingency learning effects depended on contingency awareness: Reliable long-term contingency learning effects were obtained only for those words for which the color-word contingency was correctly identified at the end of the experiment. In sum, our findings indicate that long-term color-word contingency learning reflects propositional learning.