We present a novel technique for calibrating a zooming camera based on the invariance properties of the Normalized Image of the Absolute Conic (NIAC). First, we show that the camera parameters (independent of the position, orientation and zooming) are determined uniquely by the NIAC. Then, we exploit the invariance properties to develop a stratified calibration method that decouples the calibration parameters. The method is organized in three steps: i) computation of the NIAC, ii) computation of the focal length for each image, iii) computation of the orientation and the position of the camera. The method only requires two images of non-coplanar squares at different orientations and the solution is obtained based on linear minimization that provides fast and stable convergence properties. A finer solution can be obtained by taking the solution of the linear minimisation as a starting point of a non-linear minimization. Preliminary results for synthetic and real data show that the method is capable of obtaining accurate results.