We investigate the snaking of localised patterns, seen in numerous physical applications, using a variational approximation. This method naturally introduces the exponentially small terms responsible for the snaking structure, that are not accessible via standard multiple-scales asymptotic techniques. We obtain the symmetric snaking solutions and the asymmetric 'ladder' states, and also predict the stability of the localised states. The resulting approximate formulas for the width of the snaking region show good agreement with numerical results.There has been much recent interest in the phenomenon of spatially localised patterns, [2,4,7,11], extending our understanding of earlier work on this topic [1,13]. As discussed in the review article by Dawes [9], this work is motivated by wide-ranging applications in many different areas of physics, including buckling of struts and cylinders [6,10,15], nonlinear optics [7], convection patterns [1,12], gas discharge systems and granular media. Most theoretical work concentrates on the Swift-Hohenberg equation, which is the simplest model equation that illustrates the effect [1,9,11].These localised patterns arise as a result of bistability between a uniform state and regular patterned state. A front linking these two states might be expected to move in one direction or the other, except at a specific parameter value, referred to as the Maxwell point. However, due to a pinning effect first described qualitatively by Pomeau [13], the front can lock to the pattern, resulting in a finite range of parameter values around the Maxwell point where a stationary front can exist. Combining two such fronts leads to a pinned localised state. The pinning range is seen in numerical simulations [3-5, 14, 18] and leads to a 'snaking' bifurcation diagram in which the control parameter oscillates about the Maxwell point as the size of the localised pattern increases.The localised states can also been seen from the viewpoint of spatial dynamics as a homoclinic connection from the uniform state to itself, leading to a geometrical argument for the existence of such states over a finite range of parameter values [6,8,10].It has been appreciated for some time [13] that the pinning effect cannot be described by conventional multiplescales asymptotics, since this method treats the scale of the pattern and the scale of its envelope as independent variables, leading to an arbitrary phase in the envelope function. Hence, the pinning range is 'beyond all orders', or is exponentially small in the small parameter corresponding to the pattern amplitude [1]. A very thorough analysis of the exponential asymptotics of this problem has recently been carried out by Kozyreff and Chapman [7,11], extending earlier work [15,19]. The calculation of the pinning range is extremely complicated and unfortunately requires two fitting parameters. An ansatz based on the weakly nonlinear solution is substituted into the Lagrangian of the system, and then this is minimised over the unknown parameters. This approach has a...