Structural motif analysis seeks to better understand brain connectivity by interrogating the long-range axonal projections that link distant brain regions at the single-cell resolution. This approach is enabled by MAPseq, a novel experimental methodology which labels cellular projections with viral barcodes. This recontextualizes single-neuron tracing as a problem of sequencing, overcoming the throughput limitations of prior optical techniques and capitalizing on the swift advancement of modern sequencing technology.Current MAPseq-based long range projection motif analysis methods rely on the use of standard algorithmic clustering methods. These approaches require significant transformations of the data and are not based on formal models of neural projection, limiting their biological interpretability.An alternative method overcomes these issues by modelling projection patterns via the Binomial Distribution. This method allows for proper hypothesis testing and biological interpretations but is based on a highly simplified model of neural projection which removes much of the information contained in a typical MAPseq dataset.Here we introduce a new method for MAPseq motif analysis, utilizing a novel Hierarchical Bayesian Mixture Model of neural projection based on the Dirichlet-Multinomial. This approach models neural projection directly, accommodating features of MAPseq data ignored by the Binomial Model, such as projection strength. It also does not require any transformations of the data, simplifying the biological interpretation of the results and better enabling group comparisons.The utility of this model is demonstrated here on a MAPseq dataset describing long-range projections from the Entorhinal Cortex to specific target regions in the neocortex.