Initiation of brittle fracture in welded structures was investigated by means of two newly developed testing methods : Transversely Welded Wide Plate Test and Longitudinally Welded I-Shape Test (IHI-Test). The Transversely Welded Wide Plate Test brought forth the possibility of brittle fracture along welded joint, so far believed improbable in as-welded structures. Along-weld-line brittle fracture was found occurrable in structures built from a certain type of high-strength steels. The minimum width of test plate should be such that the free edges would have no substantial effects. The mean fracture stress was found largely affected by the configuration of the joint, especially by an angular distortion, as well as by the notch toughness at the fusion zone. This fact was given theoretical explanation by calculating the strain at the fusion zone in distorted joint, and by assuming the brittle fracture to initiate when this strain amounts to a critical value, which is dependent on the material and temperature at the zone in question. Theoretical fracture stresses at given temperature of test plates, welded under various welding conditions and with different initial distortions, excellently agreed with experimental values. The Longitudinally Welded I-Shape Test reproduced low-stress brittle fracture, initiating from a natural notch in residual stress field, at considerably higher temperat res than what had been expected from existing testing methods. In this test, the width of specimen markedly affected the mean fracture stress. This phenomenon was hardly explainable by presuming as in the existing theories that the brittle fracture initiation is governed solely by the magnitude of stress at the notch front; the residual stress pattern around the notch must come upon to play a role. The anthors propose an analytical way of estimating stresses for brittle fracture initiation in such a non-uniform stress field. This analysis is essentially in line with the fracture mechanics, except that further consideration is given to the plastic zone size which should be dependent on the distribution of stresses within a certain area around the notch. The theory again excellently represented the observation. The herein presented two newly developed testing methods along with the proposed way of analysis lead to a clear-cut description of brittle fracture initiation characteristic to welded structures. Means for preventing such fracture follow at once.