In this paper, the noise sources of an airborne electromagnetic frequency domain instrument used to measure sea-ice thickness are studied. The antennas are mounted on the wings of an aircraft. The paper presents real data examples showing that strong noise limited the accuracy of the thickness measurement to +/-0.5 m in the best case. Even drift correction and frequency filtering did not reduce the noise to a level necessary for sea ice thickness measurements with an accuracy of 0.1 m. We show results of 3D finite element modeling of the coupling between transmitter and receiver coils and the aircraft, which indicate that wing flexure is the primary cause of the strong noise. Wing deflection angles below 5° relative to the fuselage are large enough to cause changes higher than the wanted signal from the seawater under the ice. Wing flexure noise can be divided into an inductive and geometric contribution, both of the same order. Most of the wing flexure signal appears on the inphase component only, hence the quadrature component should be taken for sea ice thickness retrievals when wing flexure is present even when the inphase produces a larger ocean signal. Results also show that pitch and roll movements of the aircraft and electromagnetic coupling between seawater and aircraft can contribute significantly to the total noise. For flight heights of 30 m over the ocean these effects can change the signal by about 10% or more. For highly quantitative measurements like sea-ice thickness all these effects must be taken into account. We conclude that a fixed wing electromagnetic instrument for the purpose of measurements in a centimeter scale must include instrumentation to measure the relative position of the antenna coils with an accuracy of 1/10 mm. Furthermore the antenna separation distance should be as large as possible in order to increase the measured ratio of secondary to primary magnetic field strength.