The attenuation coefficient (AC) is a tissue property that can be estimated from optical coherence tomography (OCT) data. We observed that excessive noise below the retina might cause both an underestimation and a significant variation of the estimated AC values by a state-of-the-art algorithm. Two methods were proposed to reduce these effects: I) by removing the average noise signal from the OCT data; II) by excluding the detected noise region below the retina. The methods were applied to four circular peripapillary retinal scans of a healthy subject. We evaluated all methods quantitatively using metrics for the inter- and intra-A-lines variation of the estimated ACs. Both methods resulted in higher ACs thereby reducing the bias. However, only method II succeeded in reducing the amount of variation by both metrics; method I made things worse. In conclusion, method II yields a more robust and more precise estimate of the AC, in particular for the choroid and sclera, compared to the baseline method.