Querying data in relational databases is often challenging since SQL requires its users to know the exact schema of the database, the roles of various entities in a query, and the precise join paths to be followed. On the other hand, keyword search is unable to express much desired query semantics. In this paper, we propose a query language, Schema-free SQL, which enables its users to query a relational database using whatever partial schema they know. If they know the full schema, they can write full SQL. But, to the extent they do not know the schema, Schema-free SQL is tolerant of unknown or inaccurately specified relation names and attribute names, and it also does not require information regarding which relations are involved and how they are joined. We present techniques to evaluate Schema-free SQL by first converting it to full SQL. We show experimentally that a small amount of schema information, which one can reasonably expect most users to have, is enough to get queries evaluated as if they had been completely and correctly specified.