Recently, compressive sensing (CS) theory has been applied for synthesising maximally sparse arrays, in which the best subset of sampling element locations is chosen to compose a sparse array for matching a desired radiation pattern. However, their performances are strongly depended on the proper setting of the initial sampling locations, which are typically obtained by gridding the continuous array aperture. Such a setting is usually hard to handle for large planar array synthesis. To address this problem, a precision and effective method based on the perturbed compressive sampling (PCS) is proposed. Position perturbation variables are augmented to the traditional CS‐based model, which allow continuous element placement. Then, a joint sparse recovery approach is used to optimise the excitations and position perturbations of the elements simultaneously. Moreover, the authors implement an extended PCS model with a secondary grid strategy to reduce the modelling error and the computational cost. The proposed design problem is solved with a general sparse recovery solver, named FOCal under‐determined system solver. Numerical results show that the method yields a higher array sparsity, a faster computational speed and a better pattern matching accuracy than the existing CS‐based methods.