BACKGROUND The diagnosis of well‐differentiated adenocarcinoma on bile aspiration is a well‐known challenge. This study was aimed at improving the diagnostic performance and providing a biliary cytology learning atlas. METHODS This single‐center, retrospective study included 135 cases of informative biliary samples collected between 2009 and 2018 that were classified as benign, atypical, or malignant. A double assessment was performed by a novice and a cytopathologist experienced in biliary cytology to establish the specificities, sensitivities, and inter‐ and intraobserver κ index agreements of 24 cytological criteria, which were illustrated in a learning atlas. RESULTS A multivariate logistic regression was used to assess whether the most specific and reproducible criteria were associated with malignancy. A scoring system was statistically determined: 6 points were attributed in the presence of a 3‐dimensional (3D) cluster, anisonucleosis, and a nuclear to cytoplasmic (N:C) ratio > 0.5, whereas 4 points were given in the presence of an enhanced nuclear membrane. A score higher than 10 points resulted in a malignancy diagnosis with 96% sensitivity and 97% specificity. CONCLUSIONS A diagnostic tree of malignancy based on 4 criteria, together with a multidisciplinary approach, allows the diagnosis of adenocarcinoma with a specificity of 100% and a sensitivity of 88% or 72% depending on the presence of a single malignant cell or the presence of 3 combined criteria (a 3D cluster, anisonucleosis, and an N:C ratio > 0.5). It comes with a learning atlas useful for cytopathologist training and accuracy in this uncommon cytology.;