There has recently been a surge of interest in relational database mining that aims to discover useful patterns across multiple interlinked database relations. It is crucial for a learning algorithm to explore the multiple inter-connected relations so that important attributes are not excluded when mining such relational repositories. However, from a data privacy perspective, it becomes difficult to identify all possible relationships between attributes from the different relations, considering a complex database schema. That is, seemingly harmless attributes may be linked to confidential information, leading to data leaks when building a model. Thus, we are at risk of disclosing unwanted knowledge when publishing the results of a data mining exercise. For instance, consider a financial database classification task to determine whether a loan is considered high risk. Suppose that we are aware that the database contains another confidential attribute, such as income level, that should not be divulged. One may thus choose to eliminate, or distort, the income level from the database to prevent potential privacy leakage. However, even after distortion, a learning model against the modified database may accurately determine the income level values. It follows that the database is still unsafe and may be compromised. This paper demonstrates this potential for privacy leakage in multi-relational classification and illustrates how such potential leaks may be detected. We propose a method to generate a ranked list of subschemas that maintains the predictive performance on the class attribute, while limiting the disclosure risk, and predictive accuracy, of confidential attributes. We illustrate and demonstrate the effectiveness of our method against a financial database and an insurance database.