Mouse models are attractive from a genetic and molecular standpoint for studying disease onset, progression and treatment. Longitudinal tracking of mouse brain function is best done by non-invasive and repeatable methods, such as fMRI. However mouse fMRI has many technical challenges, such as local susceptibility artifacts due to relatively large air-tissue interfaces, and the difficulty of maintaining a stable physiological state and neurovascular coupling in anesthetized or sedated mice. The aim of this study was to assess the long-term reproducibility of stimulation fMRI (stim-fMRI) and resting state fMRI (rs-fMRI) in the sedated mouse.Experimental design: We repeated scans either 3, 6 or 9 weeks apart on C57BL/6 male mice (n=7), with age at 11 weeks at time of the first scan. We used a 9.4T scanner (Agilent Technologies, USA), under a protocol optimized for mouse resting-state fMRI (0.1mg/kg/h medetomidine i.p. [1]). fMRI of both resting state and forepaw stimulation (6Hz, 0.5mA) was performed. MRI: BOLD EPI: TR/TE=2000/15ms, α=90°, slices=15, voxel=0.31x0.31x0.5mm³, volumes (resting state / forepaw stimulation) = 300/150. FSE: TR/TE=2500/40ms, echo-train=8, average=2, voxel=0.08x0.08x0.15mm³. Analyses: Postprocessing: Data were motion corrected with SPM. Ventricle and muscle signals were regressed out to avoid physiological noise. All data were coregistered onto an anatomical template using FSL FLIRT and FNIRT, followed by spatial smoothing with Gaussian kernel of 2 pixels FWHM. Resting state fMRI: Data were highpass-filtered at 0.01 Hz. Whole-brain connectivity maps were generated by seed-based correlation of averaged regions based on anatomical template and transformed using Fischer’s z-transformation. Forepaw stimulation fMRI: GLM was performed to get BOLD activation maps with threshold of p<0.01. Number of activated voxels and % signal change were calculated. Reproducibility: We assessed absolute agreement of single measures using the intraclass correlation based on a 2-way mixed model (ICC 3,1) [2]. The following metrics were evaluated: S1 forelimb and CPu interhemispheric functional connectivity z-scores, plus activated voxel counts and signal change amplitudes from the forepaw stimulation. Temporal SNR (tSNR) was calculated to assess the quality of the EPI scans across time.Both sessions for each mouse were on average 6.9 weeks apart. There was no correlation of the functional connectivity or stim-fMRI results with time span between sessions. The ICC was 0.46, p=0.15 for interhemispheric functional connectivity in S1FL region (see Fig. 1(a,c)). The ICC was 0.65, p<0.05 for interhemispheric functional connectivity in CPu (see Fig. 1(b,d)). For stim-fMRI, the ICC was 0.7, p<0.05 for BOLD activated voxel counts in S1FL region (see Fig. 2(a,c)), while BOLD signal amplitudes had an ICC of 0.68, p<0.05 (see Fig. 2(b,d)). Fig. 3 shows the tSNR plotted against: (a) interhemispheric functional connectivity in S1FL and CPu, and (b) activated voxel counts and BOLD signal amplitudes from the stim-fMRI. There was no significant correlation between tSNR and the functional connectivity scores, as well as with the stim-fMRI results (p>0.3), though a slight positive trend can be seen for functional connectivity.This is the first test-retest study on the reliability of mouse fMRI to our knowledge. Within a span of 3-9 weeks, the forepaw stimulation results had significant absolute agreement. The functional connectivity results were slightly more variable. Interhemispheric connectivity between a subcortical area like the CPu had significant reproducibility, while the cortical area S1 forelimb (S1FL) did not achieve significant reproducibility within the sample size of n=7. Nonetheless its ICC of 0.46 would still be within the range of “fair” (ICC=0.40-0.59), according to the guidelines stated in [3]. Length of time between sessions and tSNR was not significant factors in the reproducibility. For the observed variability in the functional connectivity of the cortical S1FL region, possible reasons include susceptibility artifacts and distortion around the surface, leading to coregistration mismatch, especially for a small brain region like the S1FL. On the other hand, the coregistration of the much larger CPu region is more robust, hence allowing better reproducibility of functional analysis.