BACKGROUND: Decision analytic software is commonly used to estimate long-term costs or effects of treatment by using Markov models. Monte Carlo simulation is used to estimate confidence limits (CL) for costs or effects in decision models. Exact cost or effects may be calculated by matrix inversion. CL for costs and effects may be calculated from matrix inversion by using distributions for each transition probability. OBJECTIVES: The objective of this study was to validate decision analytic software by comparing the bias in decision analytic CL and matrix CL. METHODS: A 3-state Markov model was developed. Transition probabilities, utilities, and costs were assumed. Costs and effects were discounted at 3%. Decision analytic CL and matrix CL were calculated as the interquartile range (IQR) from 10,000 simulations. For each simulated value, over or under was defined relative to the exact value. Bias was defined as the ratio of (??over – ??under)/ (??over + ??under). DATA software was used for decision analysis; S Plus was used for matrix programming. RESULTS: Estimated life years, quality-adjusted life years (QALY), and costs ($) are summarized below: Data Analytic Matrix Inversion Median (IQR) Bias Median Bias Life Years $ $ per Life Year $ per QALY 5.187 (4.812; 5.611) 3.960 (3.669; 4.286) 134,900 (110,900; 160,800) 25,900 (20,900; 31,500) 33,800 (27,400; 41,300) 0.086 0.006 -0.038 5.186 (4.803; 5.605) 3.927 (3.639; 4.251) 133,800 (109,700; 160,000) 25,900 (21,700; 30,300) 34,100 (28,200; 40,600) 0.068 -0.014 -0.017 CONCLUSIONS: Decision analytic software may yield biased estimates of costs and effects. The implications of this must be considered. Analysts and policymakers should carefully validate all decision models prior to using them to determine health policy.