Various regression models and data windows have been used to examine the multiple-specimen J integral analysis to provide the best estimate of fracture toughness. The linear regression model (similar to ASTM E813-81) results in the least accuracy in the data fitting and is highly sensitive to the data window. The power law model (similar to ASTM E813-87) gives a better data fitting and is less sensitive to the data window. The polynomial regression models with higher orders (n = 3 and 4) give the best data fitting. True crack initiation occurs gradually rather than as an abrupt event, and therefore the critical J strictly based on initiation could be in error. ASTM E813-81 based on the theoretically predicted blunting line tends to give an underestimated J(c). The power law approach of ASTM E813-87 indeed provides a better data fitting but suffers the disadvantage of an overestimated J(c) owing to the 0.2 mm offset line. The 0.1 mm offset line approach proposed here results in the most consistent J(c) regardless of the regression models and data windows selected.