The purpose of this study was to scale attacking performance in soccer games, and to confirm the causal structure model of attacking skill. The observation materials were 156 attacks which reached the shooting or assisted-pass stage during the four men's quarter final soccer games at the Sydney Olympics in 2000. Attacking game performances were assessed from the distances between attackers and defenders, and the number of defenders, using interval scales. The scaling procedure for attacking performance involved 1) analyzing the qualitative structure of attacking performances using qualitative cause and effect analysis and the Delphi method, 2) testing for objectivity, 3) selecting measurement items using reliability testing and exploratory factor analysis (EFA), 4) testing of construct validity using confirmatory factor analysis (CFA), 5) testing of factor invariance for the CFA model using multi-group analysis, 6) examination of the causal structure model using structural equation modeling (SEM) with the multiple-indicator model, and 7) examination of factor invariance of the causal structure model using multi-group analysis. From the CFA, high validity coefficients for the 3 attacking skills and 8 items were obtained. The causal structure model of attacking skill was found to be statistically valid. The highest goodness-of-fit indices were obtained in the strong factorial invariance level from multi-group analysis of the CFA model and the multiple-indicator model. It was concluded that attacking skill consisted of 3 sub-domains that could be measured from 8 game performance items in the attacking phases of soccer games.