Seven clinical sites compiled data from 4638 women who were referred directly to coronary angiography (catheterization-first strategy; n = 3375) or who underwent stress myocardial perfusion imaging (MPI) first (n = 1263) followed by coronary angiography if at least one reversible myocardial perfusion abnormality was detected. The study examines the cost minimization potential of these available invasive and noninvasive diagnostic strategies in women with chest pain.Women in both groups were subclassified by the core laboratory as being at low (<0.15), intermediate (0.15 to 0.60), or high (>0.60) pretest likelihood for coronary artery disease (CAD). Among the catheterization-first patients, at least one coronary stenosis >70% was present in 13% of low likelihood patients, 29% of intermediate likelihood patients, and 52% of patients with high CAD likelihood. Perfusion abnormality rates in the MPI-first group were 23% in low likelihood patients, 27% in intermediate likelihood patients, and 34% in high CAD likelihood patients. Of the MPI-first subset, 50%, 55%, and 76%, respectively, underwent catheterization in at least one coronary stenosis >70%. Cardiac death rates ranged from 0.5% to 2.2% in patients with CAD and did not differ from the 2 testing strategies (P = not significant). The composite cost per patient of diagnostic testing plus follow-up medical care over a period of 2.5 +/- 1.5 years (calculated for both strategies from inflation-corrected Medicare charges, adjusted for institutional cost-charge ratios) ranged from $2490 for patients with low likelihood to $3687 for patients with high likelihood with the catheterization-first strategy and from $1587 to $2585 for patients undergoing MPI first (P < .01 between risk subsets and strategies).In women referred for diagnostic evaluation of stable chest pain, MPI followed by selective coronary angiography in patients with at least 1 perfusion abnormality minimizes the near-term composite cost per patient compared with a direct catheterization-first strategy, regardless of pretest CAD likelihood.