Large arm length imbalance fibre-based interferometers have shown great potential for laser frequency stabilization and control, with frequency noise power spectral density close to the 10 -1 Hz/Hz 1/2 level [1]. In order to understand the performance of such frequency stabilization systems, it is important to distinguish the intrinsic noise of the frequency reference, which is ultimately limited by fibre thermal noise, and the locking noise, which is limited by the detection noise but can also be affected by residual amplitude modulation fluctuations in phase modulator or other less well known causes. It is often difficult to measure the out-of-loop locking noise when one laser is locked to a frequency reference. However, the fibre length noise practically cancels when two lasers are locked onto the same interferometer with a small frequency difference, which allows to measure the out-of-loop locking noise to a very low level [2]. Fig. 1 (a) shows our experimental setup. Two low frequency noise diode lasers at 1,5 m from RIO, with frequency difference in the microwave range, are independently locked on the same Michelson interferometer with a 300 m fibre spool in one arm, using the Pound-Drever-Hall (PDH) technique. The two PDH RF signals are detected by the same photodiode followed by a shot noise limited home-made transimpedance amplifier. The photodetector signal is split, filtered and down-mixed to generate two error signals which are amplified by a servo circuit with 4 integrators in order to reach very low residual laser noise at 1 kHz despite a bandwidth below 100 kHz. Frequency corrections are applied through the diode current and the electro-optic voltage. The beat-note frequency noise spectrum was measured by using a home-made frequency-to-voltage convertor. Fig. 1 (b) shows the spectrum of the frequency noise between the free running (dark) and the locked lasers (blue), the frequency noise converted from the error signals (red) and the measurement floor (grey). We clearly observe on the spectrum of the frequency noise between the two locked lasers, i.e. the out-of-loop locking noise, a white floor of 210 -1 Hz/Hz 1/2, largely above the in-loop error noise floor. Multiple tests performed on the setup have shown that this noise floor is independent of the frequency difference between the two lasers, independent of the detection power level, on the way correction was applied and also on different interferometer configurations and was not due to any cross-talk or multimode effects. However using two fibre lasers with a frequency noise lower than the RIO laser, the noise floor have decreased to 910 -2 Hz/Hz 1/2 (corresponding to a Lorentzian linewidth of 25 mHz). Moreover, adding a single tone frequency modulation at a frequency between 10 kHz and 100 kHz on one of the laser was shown to strongly increase the noise floor level as a function of the modulation amplitude. We conclude that this noise floor is due to a down conversion of the residual laser frequency noise, located here mainly between 10 kH...