Testing for PD-L1 expression by immunohistochemistry (IHC) is used to predict immune checkpoint blockade (ICB) benefit but has performed inconsistently in urothelial cancer (UC) clinical trials. Different approaches are used for PD-L1 IHC. We analyzed paired PD-L1 IHC data on UC samples using the SP142 and 22C3 assays from the phase 3 IMvigor130 trial and found discordant findings summarized by four phenotypes: PD-L1 positive by both assays (PD-L1 double positive; PD-L1DP), PD-L1 positive by the SP142 assay only (SP142 single positive; SP142SP), PD-L1 positive by the 22C3 assay only (22C3 single positive; 22C3SP), and PD-L1 negative by both assays double negative (PD-L1 double negative; PD-L1DN). PD-L1DP and SP142SP UCs were associated with more favorable ICB outcomes and increased dendritic-cell (DC) infiltration. SP142 PD-L1 staining co-localized with DC-LAMP, a DC marker, while 22C3 staining was more diffuse. 22C3SP UCs, associated with worse outcomes, were enriched in tumor cell-dominant PD-L1 expression. Multiplex IHC in an independent ICB-treated cohort confirmed that tumor cell-dominant PD-L1 expression was associated with shorter survival. Using different PD-L1 assays, we uncovered that SP142 may preferentially stain PD-L1-expressing DCs, key to orchestrating antitumor immunity, while tumor cell-dominant PD-L1 expression, which underlies a subset of "PD-L1 positive" specimens, is associated with poor ICB outcomes.