One important debate about self-compassion is whether it should be conceptualized as one general construct or two separate constructs: compassionate self-responding (CS) and uncompassionate self-responding (UCS). However, our understanding of self-compassion has been restricted to examining it as a trait-like construct assessed at a single time point, thus limiting our understanding of potentially important fluctuations in compassion within-person and over time. Therefore, to move forward the debate, the present study used a novel method, Dynamic Structural Equation Modeling, to examine the within-person temporal dynamics between CS and UCS. Participants were 114 German adults who reported their momentary state of CS and UCS six times a day for 7 days using a smartphone. Both autoregressive effects (i.e., CS at previous moment predicts subsequent CS) and cross-lagged effects (i.e., CS at previous moment predicts subsequent UCS) were modeled. We found evidence for positive autoregressive effects of both constructs while no evidence for cross-lagged effects. We further found that these effects are not constant across individuals, as indicated by the significant random effects. Our results highlight the utility of separating self-compassion into CS and UCS at the within-person level in order to better inform the development of tailored interventions. (PsycInfo Database Record (c) 2025 APA, all rights reserved).