Dynamic social networks can be conceptualized as sequences of dyadic interactions between individuals over time. The relational event model has been the workhorse to analyze such interaction sequences in empirical social network research. When addressing possible unobserved heterogeneity in the interaction mechanisms, standard approaches, such as the stochastic block model, aim to cluster the variation at the actor level. Though useful, the implied latent structure of the adjacency matrix is restrictive which may lead to biased interpretations and insights. To address this shortcoming, we introduce a more flexible dyadic latent class relational event model (DLC-REM) that captures the unobserved heterogeneity at the dyadic level. Through numerical simulations, we provide a proof of concept demonstrating that this approach is more general than latent actor-level approaches. To illustrate the applicability of the model, we apply it to a dataset of militarized interstate conflicts between countries.