Analysis of responses to L-3,4-dihydroxyphenylalanine (L-DOPA) in hemiparkinsonian rodent models can enhance understanding of Parkinson's disease (PD). Traditionally, these responses have been classified into medication OFF and ON states, with ON defined as the period after sufficient time has elapsed following drug administration. However, this binary classification may lead to neurophysiological analyses that overlook interindividual variability in the onset and duration of therapeutic effects. Although measuring plasma L-DOPA concentrations has been proposed to quantify this variability, technical difficulties in direct measurement limit the application of this approach. Here, we investigated whether acute L-DOPA responses can be quantitatively distinguished into behavioral states without direct plasma measurements, and whether these states correspond to neurophysiological dynamics. Using quantitative analysis of rotation direction in hemiparkinsonian rat model, we classified L-DOPA responses into four temporally distinct states: OFF, early contralateral rotation (ECR), rotational pause (RP), and late contralateral rotation (LCR). These behaviorally defined states exhibited statistically distinct neurophysiological profiles, characterized by simultaneously measured subthalamic firing rates and striatal dopamine levels. Although single-unit activity was insufficient to decode state transitions, neural population activity enabled robust decoding via a support vector machine (SVM). Furthermore, linear dynamical system modeling revealed transition-specific trajectories in a low-dimensional state space, suggesting that state transitions are represented by coordinated population dynamics. These findings demonstrate that temporal variability in L-DOPA responses can be identified behaviorally, providing a neurophysiologically validated framework that extends the conventional binary classification into a multi-state transition for PD research.