This study compares the effectiveness of a learning-by-doing methodology against traditional classroom-based instruction in ESP training for hospitality employees. A non-equivalent-group quasi-experimental pre-test–post-test design was implemented with 36 Food and Beverage staff members (B1 CEFR level) from a hotel chain operating in Gran Canaria, La Palma, Tenerife, and Fuerteventura (Canary Islands), allocated to a traditional instruction group (n = 18) or a learning-by-doing group (n = 18) based on staff availability. Both groups completed a 12-hour intervention delivered over three days. Performance was assessed using a researcher-developed 30-item written knowledge and situational-judgement test administered before and after the intervention. Given a significant baseline imbalance between groups, an analysis of covariance (ANCOVA) was used to compare post-test scores while controlling for pre-test performance. The learning-by-doing group achieved a higher unadjusted post-test score than the traditional group (6.91 ± 1.83 vs. 4.83 ± 2.45, p = 0.0069, d = 0.64); however, after adjusting for baseline differences, the between-group difference was no longer statistically significant (adjusted difference = 0.43, 95% CI [-0.73, 1.58], p = 0.459, partial η² = 0.017), and absolute change scores were comparable between groups (2.23 vs. 2.46 points, p = 0.6585). These findings offer preliminary, context-specific support for the potential benefits of experiential instruction in hospitality ESP, while underscoring the need for larger, better-balanced samples and direct observation of workplace performance before firm pedagogical recommendations can be made.