Mutations in Candida glabrata FKS genes, which encode the echinocandin target enzyme, are independent risk factors for treatment failures during invasive candidiasis. We retrospectively compared the ability of caspofungin susceptibility testing methods to identify C. glabrata FKS mutant isolates and predict outcomes among patients at our center. Eight percent (10/120) of sterile-site C. glabrata isolates harbored FKS1 (n = 3) or FKS2 (n = 7) mutations, including 32% (10/32) recovered from patients with prior echinocandin exposure. Median echinocandin exposures for mutant and nonmutant isolates were 55 (range, 7 to 188) and 13 (3 to 84) days, respectively (P = 0.004). Sensitivity and specificity of the CLSI caspofungin resistance breakpoint MIC (>0.12 μg/ml by broth microdilution using RPMI medium [BMD-RPMI]) were 90% (9/10) and 3% (3/110), respectively, for identifying FKS mutants. Sensitivity and specificity of receiver-operator characteristic (ROC) curve-derived breakpoints by BMD-RPMI, BMD-antibiotic medium 3, Etest, and YeastOne ranged from 70 to 100% and 89 to 95%, respectively; susceptibility rates varied from 83 to 90%. The 14-day echinocandin treatment success rate was 67% (44/66); failure was more likely with prior echinocandin exposure (P = 0.002) or infection with an FKS mutant (P = 0.0001) or echinocandin-resistant isolates by BMD-AM3, Etest, and YeastOne (P ≤ 0.03). The failure rate among patients with prior exposure and infection with a resistant isolate was 91% (10/11); it was 22% (12/55) among others (P < 0.0001). In conclusion, ROC-derived caspofungin MIC breakpoints by several methods were sensitive and specific for identifying C. glabrata FKS mutant isolates. Mutations were seen exclusively among patients with prior echinocandin exposure. A paradigm that considers prior echinocandin exposure and caspofungin MICs accurately classified treatment outcomes for C. glabrata invasive candidiasis.