Internal medicine (IM) residents and hospitalist physicians commonly perform thoracenteses. National data show that thoracenteses are also frequently referred to other services such as interventional radiology (IR), increasing healthcare costs. Simulation-based mastery learning (SBML) is an effective method to boost physicians’ procedural skills and self-confidence. This study aimed to (1) assess the effect of SBML on IM residents’ simulated thoracentesis skills and (2) compare thoracentesis referral patterns, self-confidence, and reasons for referral between traditionally trained residents (non–SBML-trained), SBML-trained residents, and hospitalist physicians. A random sample of 112 IM residents at an academic medical center completed thoracentesis SBML from December 2012 to May 2015. We surveyed physicians caring for hospitalized patients with thoracenteses during the same time period and compared referral patterns, self-confidence, and reasons for referral. SBML-trained resident thoracentesis skills improved from a median of 57.6% (interquartile range [IQR] 43.3–76.9) at pretest to 96.2% (IQR 96.2–100.0) at post-test (P < 0.001). Surveys demonstrated that traditionally trained residents were more likely to refer to IR and cited lower confidence as reasons. SBML-trained residents were more likely to perform bedside thoracenteses. Hospitalist physicians were most likely to refer to pulmonary medicine and cited lack of time to perform the procedure as the main reason. SBML-trained residents were most confident about their thoracentesis skills, despite hospitalist physicians having more experience. This study identifies confidence and time as reasons physicians refer thoracenteses rather than perform them at the bedside. Thoracentesis SBML boosts skills and promotes bedside procedures that are safe and less expensive than referrals.