Abstract Purpose: Adenoid cystic carcinoma (ACC) is a heterogeneous malignancy, and no effective systemic therapy exists for metastatic disease. We previously described two prognostic ACC molecular subtypes with distinct therapeutic vulnerabilities, ACC-I and ACC-II. In this study, we explored the ACC tumor microenvironment (TME) using RNA-sequencing and spatial biology to identify potential therapeutic targets. Experimental Design: Tumor samples from 62 ACC patients with available RNA-sequencing data that had been collected as part of previous studies were stained with a panel of 28 validated metal-tagged antibodies. Imaging mass cytometry (IMC) was performed using the Fluidigm Helios CyTOF instrument and analyzed with Visiopharm software. The B7-H4 antibody–drug conjugate AZD8205 was tested in ACC patient-derived xenografts (PDX). Results: RNA deconvolution revealed that most ACCs are immunologically “cold,” with approximately 30% being “hot.” ACC-I tumors with a poor prognosis harbored a higher density of immune cells; however, spatial analysis by IMC revealed that ACC-I immune cells were significantly restricted to the stroma, characterizing an immune-excluded TME. ACC-I tumors overexpressed the immune checkpoint B7-H4, and the degree of immune exclusion was directly correlated with B7-H4 expression levels, an independent predictor of poor survival. Two ACC-I/B7-H4-high PDXs obtained 90% complete responses to a single dose of AZD8205, but none were observed with isotype-conjugated payload or in an ACC-II/B7-H4 low PDX. Conclusions: Spatial analysis revealed that ACC subtypes have distinct TMEs, with enrichment of ACC-I immune cells that are restricted to the stroma. B7-H4 is highly expressed in poor-prognosis ACC-I subtype and is a potential therapeutic target.