Presented at the 2014 Aquarius/SAC-D Science Team MeetingRadio Frequency Interference (RFI) is a well documented issue for microwave radiometry, and L-band radiometer in particular. To mitigate against RFI, the Aquarius radiometer processing using a time-based filter, where observations that are greater than about 4-sigma above the noise floor are tagged as being RFI and removed. This filter is applied to the 10ms Aquarius observations prior to averaging to the baseline 1.44 second sampling. This algorithm works well for strong RFI sources, but is blind to any low-level RFI that is below the 4-sigma level above the noise. This case occurs frequently over the ocean within about 1000km of land, where RFI enters from many directions through the antenna pattern sidelobes. The antenna pattern reduces the magnitude of interference by typically 30dB, but when the sources combine, it is not uncommon for these sources to bias the radiometer TBs by 0.5 K, which would not be detectable with the RFI mitigation algorithm. This paper describes an algorithm to remove this low-level RFI which can be applied to future Aquarius processing versions. The algorithm uses pre-computed ascending and descending RFI contamination maps to remove the bias due to this contamination. This presentation will focus on how the RFI maps are generated. The first part is identifying the location of the sources and the second part is identifying their strength. This is done iteratively by minimizing a cost function between the observed ascending-descending TB differences for each horn and a forward model of the Aquarius antenna patterns convolved with the pre-identified RFI source locations. The presentation will provide an estimate for the algorithm performance and residual errors.