Multi‐center kidney paired donation (KPD) is an exciting new transplant option that has not yet approached its full potential. One barrier to progress is accurate virtual crossmatching for KPD waitlists with many highly sensitized patients. Virtual crossmatch results from a large multi‐center consortium, the National Kidney Registry (NKR), were analyzed to determine the effectiveness of flexible center‐specific criteria for virtual crossmatching. Approximately two‐thirds of the patients on the NKR waitlist are highly sensitized (>80% CPRA). These patients have antibodies against HLA‐A (63%), HLA‐B (66%), HLA‐C (41%), HLA‐DRB1 (60%), HLA‐DRB3/4/5 (18–22%), HLA‐DQB1 (54%) and HLA‐DPB1 (26%). With donors typed for these loci before activation, 91% of virtual crossmatches accurately predicted an acceptable cell‐based donor crossmatch. Failed virtual crossmatches were attributed to equivocal virtual crossmatches (46%), changes in HLA antibodies (21%), antibodies against HLA‐DQA (6%), transcription errors (6%), suspected non‐HLA antibodies (5%), allele‐specific antibodies (1%) and unknown causes (15%). Some failed crossmatches could be prevented by modifiable factors such as more frequent assessment of HLA antibodies, DQA1 typing of donors and auditing data entry. Importantly, when transplant centers have flexibility to define crossmatch criteria, it is currently feasible to use virtual crossmatching for highly sensitized patients to reliably predict acceptable cell‐based crossmatches.