Ensuring that modern systems comply with privacy regulations, like GDPR’s purpose limitation and user consent, is complex, time-consuming, and error-prone. Recently, a model-driven privacy approach has been proposed, enabling developers to declaratively specify a privacy model from which correct enforcement code is automatically generated. This approach promises to reduce software development complexity and minimize developer errors. We report on the results of an empirical study with 61 participants comparing model-driven and traditional code-centric approaches to implementing privacy requirements in web applications. The participants were graduate computer science students with relevant software engineering and privacy training, making them good developer proxies. They were tasked with implementing the same web application using both approaches. We evaluated the resulting implementations for correctness, development effort, and maintainability. Our findings provide empirical evidence of the superior efficiency and correctness of the model-driven approach. Despite limited prior experience, participants using the model-driven approach wrote eight times less code and produced applications with 64% fewer privacy violations than the participants using the code-centric approach. This is the first study to empirically analyze and demonstrate the benefits of model-driven approaches for developing privacy-critical applications.