Whereas mechanisms underlying lexical access and semantic priming have been widely investigated, neurocognitive processing of specific semantic relations remains controversial. Previous studies on relational semantic priming have found that related word pairs (e.g., long- short) show stronger priming effects if they are preceded by other prime-target pairs that share the same type of semantic relationship (consistent pairs) compared to pairs of a different semantic relation (inconsistent pairs). In event-related potentials (ERPs), this consistency effect was associated with a reduced negative amplitude in the late N400 time window, and its long latency of 400-700 ms after target word onset was taken to reflect post-lexical integration processes in working memory (Steinhauer et al., 2017). However, given the short 250 ms SOA in that study, it was not possible to rule out delayed prediction-based effects. Moreover, syntactic word category changes for inconsistent pairs (e.g., adjectives within a list of noun pairs) were not controlled for. In the present ERP study, we increased the SOA to 450 ms and recorded EEG while 40 native speakers of French read and evaluated 720 word-pairs nested into 9 experimental lists, each promoting a specific semantic relation and a specific syntactic word category. We found ERP consistency effects starting early on the prime word, as well as N250 and N400 effects reflecting effects on both the form and meaning processing of the target word. Thus, word pairs with a list-inconsistent semantic relation displayed smaller priming effects than consistent word pairs, and priming was even more reduced when pairs also had a different syntactic category. Finally, participants with faster response times for related pairs displayed stronger effects on the prime word, reflecting proactive and strategic processing. In contrast, slower participants displayed smaller or no effects on the prime, but instead late positivities after target onset that could be associated with reactive processing. These results suggest that predictive mechanisms are likely to underlie optimal relational semantic priming. However, prediction of the target word is not systematic, and relational priming can also happen at a slower pace, based on an evaluation of the semantic relation that follows lexical access.