Fiber waviness and ply stacking errors compromise the mechanical performance of multi-layer composites. Measuring the local in-plane fiber orientation layer-by-layer is therefore important to assure the structural integrity of a composite. The Radon transform and the Gabor filter methods have been reported to be appropriate for extracting the fiber orientation from an ultrasonic dataset, where the Radon transform method was found to be more accurate and stable. This paper proposes a tomographic reconstruction method on the basis of classical ultrasound pulse-echo testing, in view of reconstructing the local in-plane fiber orientation in multi-layer composites. A 3D ultrasonic dataset is first sliced successively through depth, and each 2D slice is locally analyzed by a Gabor filter method coupled to the concept of an Information Diagram (GF-ID). The GF-ID improves the Gabor filter method by constructing local Gabor filters with optimal orientation and wavelength from which the local in-plane fiber orientations can be obtained in an automated way. The performance of the developed GF-ID method is investigated on both synthetic texture images and an experimental ultrasonic dataset obtained from a 24-layer [45/0/-45/90]3S carbon fiber reinforced composite. It is found that the GF-ID method yields a high-quality tomographic reconstruction of the local orientations at different scales, and shows high noise resistance. Comparison with the classical Radon transform approach reveals the higher performance of the GF-ID method for ultrasonic reconstruction of the local in-plane fiber orientation in multi-layer composites.