There are many techniques for locating an object or person in an indoor environment: fingerprinting is one of the most widely used for its simplicity and excellent results. This technique uses radio frequency signals, such as Bluetooth Low Energy (BLE) or Wi-Fi and, in a calibration phase, involves creating a “fingerprint” of each location of the environment based on the signal strength, and then using that information to determine the location of a device or person in a positioning phase using, for example, a machine learning algorithm such as WKNN. This technique is not exempt from producing large errors in some specific cases, even if the accuracy is low, because the main assumption of the fingerprinting method is not always true: close points in signal or feature space must be spatially close points. There are many sources of large errors, such as different calibration-positioning point pairs receiving signals from a number of different emitters, affecting signal comparison metrics. Using an RSS database from a previous work by the authors, this study analyzes those sources of large errors in the fingerprinting method for BLE. As a first approximation to a generalization of the corrections of these large errors, a 50% reduction of the point error has been achieved in some specific cases, such as “penalizing” in the WKNN algorithm those reference points far from target points, according to the distances to the emitters.