An approach for the direct design of LPV controllers from data is proposed. This approach, called Direct FeedbacK (DFK) design, overcomes some relevant problems typical of the standard design methods, such as model uncertainty and complex control design and/or implementation. After a closed-loop stability analysis, an algorithm for direct control design from data is presented, completely based on convex optimization. The resulting controller is able to provide closed-loop stability when the number of data used for design is sufficiently large and is almost-optimal (in a worst-case sense) in terms of tracking accuracy.