Internet addiction (IA) is widespread, comorbid with other conditions, and commonly undetected, which may impede recovery. The Internet Addiction Test (IAT) is widely used to evaluate IA among healthy respondents, with less agreement on its dimensional structure. This study investigated the factor structure, invariance, predictive validity, criterion validity, and reliability of the IAT among Spanish women with eating disorders (EDs, N = 123), Chinese school children (N = 1072), and Malay/Chinese university students (N = 1119). In school children, four factors with eigen values > 1 explained 50.2% of the variance, with several items cross-loading on more than two factors and three items failing to load on any factor. Among 19 tested models, CFA revealed excellent fit of a unidimensional six-item IAT among ED women and university students (χ2(7) = 8.695, 35.038; p = 0.275, 0.001; CFI = 0.998, 981; TLI = 0.996, 0.960; RMSEA = 0.045, 0.060; SRMR = 0.0096, 0.0241). It was perfectly invariant across genders, academic grades, majors, internet use activities, nationalities (Malay vs. Chinese), and Malay/Chinese female university students vs. Spanish women with anorexia nervosa, albeit it was variant at the scalar level in tests involving other EDs, signifying increased tendency for IA in pathological overeating. The six-item IAT correlated with the effects of internet use on academic performance at a greater level than the original IAT (r = −0.106, p < 0.01 vs. r = −0.78, p < 0.05), indicating superior criterion validity. The six-item IAT is a robust and brief measure of IA in healthy and diseased individuals from different cultures.