Electrical capacitance tomography (ECT) has recently been developed for the in-situ measurement of the cross-sectional gas-solid distribution in a duct carrying a non-conductive 2-phase flow. The technique rehes on the detection of electrical capacitances between electrodes placed on the periphery of the duct. The data depend on the permittivity distribution inside the cross-section, which has to be reconstructed. Information on the flow regime, vector velocity, and gas-sohd distribution in process vessels and pipehnes can be determined from the reconstructed images. The main bottleneck is still the poor quality ofthe image reconstruction part. In the frame ofthe ECT project at Delft University of Technology (DUT), research is done towards a new direct reconstruction algorithm, based on the least squares solution. The investigation has two goals. The investigation ofthe influence of noise on the data and the reconstruction of the permittivity distribution from measured data using well known material distributions as test cases to determine the spatial resolution. To investigate the influence of noise on the reconstruction, simulated noise is added to synthetic data. The reconstructions show that the more constraints are added to the least squares solution, the more robust the algorithm becomes. Noise measurements show that the noise level of real measured data has a negligible influence on the constrained least squares reconstruction. Reconstruction of a rod (about | of the pipe diameter) near the wall of the sensor is possible. In the centre region of the pipe the quality of the reconstructed image decreases due to the under-determinacy in the centre of the pipe. Two rods can be reconstructed when both positioned close to the pipe-wall. However, they can not be seen seperately in the centre region and when they are close to one another. Bubbles (about 1/10 of the pipe diameter) are hard to reconstruct when positioned in the pipe centre. Due to their finit height 3-D effects disturb the interpretation of the measurements. The bubbles, however, can be detected when looking at the raw data.