This paper presents a new, broadband (from 0.2 mHz to 18 Hz) estimate of the self‐noise of the Streckeisen STS‐2 sensor based on the application of the three‐channel correlation technique (Sleeman et al., 2006) to one year of continuous seismic recordings. A novel presentation of instrumental self‐noise is shown using a probability density function. The measurements were obtained from three collocated sensors in the Conrad Observatory (Austria); the sensors are covered with a new type of thermal insulation consisting of thin layers of neoprene, which significantly reduces noncoherent signals between the sensors. We observe a noise reduction ranging from about 10 dB at 1 mHz to 2 dB at 1 Hz when compared to noise estimates from noninsulated sensors. The new self‐noise estimate for the STS‐2 falls below the new low‐noise model (NLNM; Peterson, 1993) at frequencies above 4 mHz and even reaches the NLNM at frequencies between 0.2 mHz and 4 mHz during optimal conditions. From our observations, we conclude that the broadband STS‐2 sensor can achieve a noise power level of 1.56×10−18 m2/s4/Hz, or −178 dB, at 1000 s. We also show results from synthetic experiments to quantify the effect of sensor misalignment on the self‐noise estimate and to quantify the misalignment between the sensors in our Conrad experiment. The synthetic experiment shows that the three‐channel correlation technique potentially can extract noise from signals with signal‐to‐noise ratio (SNR) up to 140 dB but is in practice limited to signals with 60–80 dB SNR when the alignment errors are on the order of 0.2° and 0.02°, respectively. From the observations, we conclude that the misalignment between the sensors in the Conrad experiment is on the order of 0.2°.