Background: HER2-targeted neoadjuvant chemotherapy (NAC) possesses heterogeneous outcomes and currently lacks clinically-accepted markers of response. A means of predicting which patients will benefit prior to the treatment could reduce toxicity and the delay to effective intervention. Computational analysis of MRI via a deep neural network has shown promise in identifying NAC responders among mixed receptor subtype and treatment regimen cohorts, but faces challenges due to reproducibility across institutions and has not yet been explored in the context of HER2-targeted therapy. Here we present a deep learning approach for predicting response to HER2-targeted NAC from pre-treatment MRI. Methods: 100 HER2+ breast cancer patients who received NAC with docetaxel, carboplatin, trastuzumab, and pertuzumab at Cleveland Clinic (CCF) and had pre-treatment contrast-enhanced MRI’s were included in this analysis. 49 patients achieved pathological complete response (pCR, ypT0/is), while 51 patients retained presence of residual disease following NAC (non-pCR). 85 patients were used to train a convolutional neural network to predict pCR based on pre- and post-contrast MRI images, and the model design was optimized based on performance within a 15 patient internal validation cohort. An external, held-out testing dataset consisting of 28 patients (16 pCR, 12 non-pCR) imaged and treated at University Hospitals (UH) Cleveland Medical Center was used to validate the performance of the model. Performance was assessed by area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. A multivariable model incorporating age, hormone receptor status, stage, and tumor size was developed and similarly evaluated. Results: The neural network was able to predict the response to HER2-targeted NAC in the internal validation cohort (AUC = 0.93) as well as in an independent cohort from a separate institution (AUC = 0.85). This model offered superior performance compared to a multivariate clinical model, which achieved AUC = 0.67 and AUC = 0.52, in internal validation and external held-out testing cohorts, respectively. Conclusions: Deep learning analysis of contrast-enhanced MRI could be used to better target anti-HER2 therapy by pre-treatment prediction of response.