This paper presents results of a study about the statistical properties of noise (including unknown interference) in blank channels of TV bands, i.e., TV white spaces. Measurements were conducted to collect data in three different scenarios: sunny-day laboratory, sunny-day curbside and rainy-day laboratory. Impulsive noise has been modeled as a pulse train and the pulse amplitude, pulse duration time and inter-arrival time in both time and frequency domains were defined and considered as typical statistical variables. Several comparisons were made to study the statistical properties of TV white space noise, in two different channels (672∼680MHz and 768∼776MHz), under three different scenarios, and with several typical distribution models (Rayleigh, Gamma, lognormal and Middleton A). It has been found that noises in different TV white spaces have highly consistent statistical properties, whereas under different scenarios may exhibit diverse statistical properties. A Kolmogorov-Smirnov test was applied to find the closest distributions within the considered candidate distributions. Two new models for the amplitude probability distribution of noise in the frequency domain were proposed and validated, both consisting of a low-amplitude thermal noise component and a high-amplitude impulsive noise component.