Flicking is a common interaction technique to move objects across large interactive surfaces, but little is known about its suitability for use on non-planar, curved surfaces. Flicking consists of two stages: First, visually determining the direction in which to flick the object, then planning and executing the corresponding gesture. Errors in both stages could influence flicking accuracy. We investigated flicking interactions on curved interactive surface to evaluate which type of error influences accuracy. Therefore, we carried out three user studies to analyze how each stage of flicking on a curved surface is influenced. Our main findings are: 1) Flicking gestures are more accurate if horizontal and vertical surface are joined by a continuous curve than if they are separated by an edge or gap. 2) Flicking gestures on curved surfaces are mostly influenced by the motor execution stage of the gesture rather than the visual perception stage. 3) Flicking accuracy decreases as the starting point of the gesture is moved closer to the curve. 4) We conclude with a first mathematical model to estimate the error users will make when flicking across a curve.