An accurate determination of cloud particle phase is required for retrieval of other cloud properties. The main purpose of this letter is to demonstrate that the cloud phase assumption made in previous cirrus clouds retrievals using infrared split window signatures is not relevant. Coincident observations of the Earth surface from ERS-2 and ADEOS-1 satellites give us the opportunity to compare clouds infrared signatures, as observed by ATSR-2 onboard ERS-2, with cloud thermodynamic phase derived from POLDER onboard ADEOS-1. We find out, firstly, that large Brightness Temperature Differences (BTD), estimated between 11 and 12 µm, may occured for cold liquid water clouds consisting of small supercooled droplets (i.e. altostratus). These BTDs have no longer been attributed to cirrus clouds as previous studies have been. Secondly, the probability to observe ice or water clouds with respect to their cloud top temperature is quantified for our data set. A sharp transition between the ice and water phase is shown for cloud top temperatures ranging between 240 and 260 K. The probability for a cloud to be composed of super-cooled liquid droplets is shown to be higher over land than over ocean. This points out that ice activation susceptibility is more efficient in maritime air mass than in continental one.