OBJECTIVES: To examine the association of HbA1c at diagnosis with cost and clinical outcomes in patients with type 2 diabetes (T2D) using 3 different diabetes treatment strategies.  METHODS: An existing Monte Carlo diabetes model was used to generate demographic and clinical characteristics for 10,000 random patients simulated from the time of T2D diagnosis.  Fixed initial HbA1c values were assigned (range: 7-12%).  Input distributions were derived from the literature, with diabetes complications occurring in:  retinopathy, nephropathy, neuropathy, coronary heart disease, and stroke.  The diabetes-related costs, percentage of patients reaching HbA1c target (<7%), complication events, and mortality over 10 years were evaluated using the following treatment strategies:  (S1) addition of oral anti-hyperglycemic agents (OHAs) at 3-month intervals then starting insulin after 9 months; (S2) addition of OHAs at 6-month intervals then starting insulin after 2 years; and (S3) addition of OHAs only.  All treatment strategies began with metformin then sulphonylurea and/or thiazolidinedione were added if target was not reached.  RESULTS: Diabetes-related costs increased as initial HbA1c increased for all 3 strategies.  S1 had the greatest increase ($5,300-$7,750) followed by S3 ($4,200-$5,250) per 1% HbA1c increase from 7% to 10%.  S2 had the smallest rate of increase ($3,350-$4,950).  S3 was the least costly until HbA1c exceeded 10%, however even at HbA1c of 8%, S3 had fewer patients ever reaching target (S3=79% versus S1 and S2=95% [standard errors, SE<0.41%]).  For 10,000 patients with initial HbA1c of 9%, the total counts of complications were:  S1=4,360; S2= 4,126; S3=5,009 [SE<67] with mortality rates of 42.5%, 41.9%, and 45.2% [SE <0.56%], respectively. CONCLUSIONS: In this model, S2 had the lowest complication rates and mortality in patients with T2D.  Starting HbA1c affected S1 cost more than other strategies.  Strategies with other treatments or alternative timing strategies can be specified and analyzed using this model.