(A) Illustration of first level analysis. Both for the picture viewing task pre and picture viewing task post, activity of all voxels within a ROI (e.g. bilateral hippocampus) is extracted across all trials, in which 16 different items are presented 12 times (for illustrative purposes, procedures here are depicted for 5 items only). Voxel patterns for every item in every repetition are correlated with voxel patterns for every other item in every other repetition, yielding one average cross-correlation matrix for all items, respectively for the PVT pre and the PVT post task. In the next step, the difference between the PVT post cross-correlation matrix and the PVT pre cross-correlation matrix is formed to get a difference matrix with pattern similarity increases/decreases for every item pair. This difference matrix (PS’) is then put in relation to an external variable, for example the remembered spatial distance between every item pair, which is based on the behavioral distance judgment task at the end of the experiment. The relationship between PS’ and the external variable is expressed with a correlation coefficient. For example, higher pattern similarity increases for item pairs with lower remembered distance between them (i.e. which were remembered as being closer together) will result in a negative correlation coefficient. To estimate the strength of this relationship, the correlation coefficient is compared to a distribution of surrogate correlation coefficients derived from correlating shuffled pattern similarity increases and distance judgments. The position of the real correlation coefficient in this distribution is a marker for the strength of the effect and is expressed with a z-value, whose absolute value will be higher for more extreme values with regard to the surrogate distribution. However, the z-value can be both positive and negative, depending on which tail of the distribution the real correlation coefficient is located at. (B) Second level analysis. The z-statistics from the first level analysis, which were calculated for every participant, are then tested for significance across participants by comparing the mean z across participants to surrogate mean z-values derived from averaging randomly sign-flipped first-level z-values, with 10,000 repetitions of the random sign-flips. Again, if the mean of the first-level z-values is at an extreme end of the surrogate distribution, this is reflected in a high absolute z-value and a low probability (p) that the effect is not significantly different from zero. See Figure 4—figure supplement 1 for a corresponding illustration of methodological procedure for the searchlight analysis.