Walking is known to be good for mental and physical health. However, its effects on neural activity remain unclear. To investigate the neural effects of walking, we utilized whole-brain fMRI-based predictive models (“emotion decoders”) for valence and arousal, trained with multivariate pattern analysis (MVPA). Decoder development comprised a discovery phase (model training and internal validation) and a validation phase (external cohort and confirmatory tests). Data collection for the discovery phase is complete; this preregistration specifies the validation-phase acquisition and analyses. In 2024, we completed the fMRI data collection for 32 healthy participants, forming the discovery cohort. In the discovery-phase experiment, participants completed two tasks: an emotional rating task and a one-back task. In the rating task, fMRI data were collected during picture viewing and the emotional rating process. The picture-viewing data period was used to train each valence and arousal neural decoder. Specifically, participants underwent an fMRI paradigm in which they were presented with four types of emotional pictures (pleasant-high arousal; pleasant-calm; unpleasant-high arousal; unpleasant-calm). Following each picture, participants were asked to rate the level of each valence and arousal on a continuous scale of 1 to 100 (from unpleasant [-50] to pleasant [50], and from calm [-50] to arousal [50]). For the brain data analysis, we performed pre-processing of the fMRI data using the SPM default pipeline. Specifically, functional images were realigned to the mean image of the series, slice-time corrected, motion corrected, co-registered to the structural image, normalized to MNI space, and spatially smoothed with a 6-mm FWHM Gaussian kernel. In addition, potential outlier scans were identified from the resulting subject-motion estimates and from BOLD signal indicators using default thresholds in the CONN toolbox preprocessing pipeline (5 standard deviations above the mean in the global BOLD signal change, or framewise displacement values above 0.9 mm). Then, the single-trial first-level fMRI analysis was performed to obtain beta images for each picture rating per participant. We used GLMsingle (Prince et al., 2022) to estimate single-trial beta maps per picture at the individual-subject level. After that, we applied whole-brain MVPA to obtain patterns that predict participants’ valence and arousal ratings from single-trial beta maps for each picture. The brain mask was restricted to the gray matter. For the decoding method, we employed the LASSO-PCR algorithm (Wager et al., 2013), based on individual beta maps for each picture, as features to predict participants’ emotional experience. We used a 10-fold participant cross-validation procedure to evaluate the decoder's performance. In the discovery cohort (N=32), the neural valence and arousal decoders predicted participants’ ratings with mean within-subject trial-wise Pearson correlations exceeding r = 0.45 for valence and r = 0.35 for arousal. Next, participants performed a one-back task on emotional pictures in the fMRI scanner after either walking or reading. We applied the valence and arousal decoders to fMRI data time-locked to picture viewing to obtain decoder-derived neural emotion scores. We then compared these neural emotion scores between the walking and the reading control conditions in a within-subject design. In the walk condition, participants walked on a treadmill at a self-selected comfortable pace. Whereas in the reading control condition, participants read a book while seated. After 20 minutes of walking or reading, participants entered the fMRI scanner. The order of conditions was counterbalanced across participants. During the one-back task, participants viewed a picture at a time and pressed a button when they identified a repetition of an image presented in a previous trial. The procedure ensures that their attention is sustained. To demonstrate the neural effects of walking, we fitted the neural emotion decoder on the fMRI data collected during the picture-viewing period. We obtained and compared the neural emotion score for each walking and reading condition. As a result, we found that decoder-derived neural valence scores for pleasant-high arousal pictures were higher after walking than after reading. A crucial next validation phase for pre-registration will (i) externally evaluate decoder performance and (ii) replicate and extend the findings that walking increases decoder-derived neural valence score relative to reading for pleasant-high arousal pictures. We will acquire fMRI and behavioral data from an independent validation cohort (planned N=30) under a similar experimental paradigm. The validation phase includes a one-back task and a picture-rating task identical to those in the discovery phase. In this phase, the one-back task will be performed initially. The rating task will be used to assess decoder generalization both after walking and after reading conditions (within-subject, counterbalanced). In the discovery cohort, we did not investigate whether subjective evaluations differed between walking and reading conditions, so in the validation cohort, we plan to have participants evaluate under both conditions. Furthermore, we will investigate whether the decoder's predictions hold up even when the same emotional images are shown multiple times. All preprocessing and first-level modeling will follow the analysis pipeline in the discovery phase, yielding single-trial beta maps. The valence and arousal decoders trained by the discovery dataset will be applied to the validation data. We expect that, if the neural emotion decoders in the discovery phase can be generalized to out-of-sample participants, statistically significant prediction-outcome correlations will be observed in the validation phase for individuals in the cohort. Moreover, decoder-derived neural valence scores will be compared between walking and reading conditions, with a focus on pleasant and high-arousal pictures. We expect higher neural pleasure scores and subjective pleasure ratings after walking than after reading, providing robust evidence for a neural effect of walking.