Surface roughness is known to amplify hydrophobicity. The apparent contact angle of a drop on a rough surface is often modeled using either Wenzel's or Cassie's formulas. These formulas, along with an appropriate energy analysis, are critical in designing superhydrophobic substrates for applications in microscale devices. In this paper we propose that double (or multiple) roughness structures or slender pillars are appropriate surface geometries to develop "self-cleaning" surfaces. The key motivation behind the double structured roughness is to mimic the microstructure of superhydrophobic leaves (such as lotus). Theoretical analysis similar to that presented in the paper can be used to obtain optimal geometric parameters for the rough surface. The calculation procedure should result in surface geometries with excellent water repellent properties.