A boundary layer whose thickness is of the order of the mean free path squared divided by the radius of the curvature of the boundary is proposed to exist at the bottom of the kinetic boundary layer (or Knudsen layer) when the boundary is convex in shape.