Diffusion tensor imaging (DTI)-based tractography is one of the cornerstones of the Human Connectome Project, by enabling in vivo mapping of anatomical connections within the brain [1]. However, the exact relationship between DTI-based tract representations (streamlines) and true axonal projections remains uncertain. To elucidate this relationship, we compared the accuracy of two tractography approaches in rat brain connectomics against gold standard neuroanatomical tracer data. Perfusion-fixed postmortem brains – inside the skull – of ten healthy adult (12-13 weeks) male Wistar rats were positioned in a container filled with proton-free oil (Fomblin^®). High spatial and angular resolution DTI (8-shot 3D EPI; TR/TE= 500/32.4 ms; field-of-view = 19.2 x 16 x 33 mm³; voxels = 150 x 150 x 150 µm³; 5 b₀ images; diffusion-weighting in 60 non-collinear directions with b = 3842.23 s/mm²; four averages) was conducted on a 9.4T Varian MR scanner. After DTI acquisition, images were non-rigidly aligned to a 3D model of the Paxinos and Watson rat brain atlas. Whole-brain connectomes (50.000 streamlines) were obtained for each animal with conventional diffusion tensor-based tractography and constrained spherical deconvolution (CSD)-based tractography. CSD takes into account the presence of crossing fiber populations and uses 2^nd order integration over fiber orientation distributions [2]. Tractography was repeated with different fractional anisotropy (FA) thresholds (0.15, 0.2) and varying maximum angle between tracking steps (10-90°). The overall presence or absence of connecting streamlines between specific cortical regions were compared to findings from a rat brain neuronal tracer database [3] (Figure, left). Reconstruction accuracy was expressed as sensitivity and specificity (i.e. the proportion of correctly identified present (true positive) and absent (true negative) connections; as mean ± standard deviation, at group level).Tensor-based tract reconstructions had low (between hemispheres) to moderate (within hemispheres) sensitivities to identify connections (Figure, middle). This increased with more liberal tracking angle and lower FA thresholds. Increasing sensitivity occurred at the expense of decreasing specificity. The sensitivity and specificity to identify neuronal projections with CSD were also relatively low, but much less affected by tracking angle and FA threshold (Figure, right).High-resolution DTI-based tractography in rat brains revealed a significant amount of false positive and false negative streamline connections when compared to neuroanatomical tracer data. Comparable degrees of sensitivity and specificity have been shown for standard DTI-based tractography in mice [4]. The current analysis shows that accounting for crossing-fibers – with CSD – does not significantly improve sensitivity and specificity of DTI-based reconstructed cortical projections.High spatial and angular resolution DTI-based tract reconstructions correlate incompletely with true axonal projections in rat brain, even when using an advanced fiber reconstruction algorithm such as CSD. Similar findings have been reported for rhesus monkey brain [5], suggesting that DTI-based connectomics should be interpreted with care. We suggest that anatomically constrained tractography or global tractography may increase DTI-based connectome accuracy.