T cells play fundamental roles in adaptive immunity, relying on a diverse repertoire of T-cell receptor (TCR) α and β chains. Diversity of the TCR β chain is generated in part by a random yet intrinsically biased combinatorial rearrangement of variable (Vβ), diversity (Dβ), and joining (Jβ) gene segments. The mechanisms that determine biases in gene segment use remain unclear. Here we show, using a high-throughput TCR sequencing approach, that a physical model of chromatin conformation at the DJβ genomic locus explains more than 80% of the biases in Jβ use that we measured in murine T cells. This model also predicts correctly how differences in intersegment genomic distances between humans and mice translate into differences in Jβ bias between TCR repertoires of these two species. As a consequence of these structural and other biases, TCR sequences are produced with different a priori frequencies, thus affecting their probability of becoming public TCRs that are shared among individuals. Surprisingly, we find that many more TCR sequences are shared among all five mice we studied than among only subgroups of three or four mice. We derive a necessary mathematical condition explaining this finding, which indicates that the TCR repertoire contains a core set of receptor sequences that are highly abundant among individuals, if their a priori probability of being produced by the recombination process is higher than a defined threshold. Our results provide evidence for an expanded role of chromatin conformation in VDJ rearrangement, from control of gene accessibility to precise determination of gene segment use.