Seismic interferometry is a technique to retrieve Green’s functions between two points from crosscorrelation functions of seismic ambient noise records at the two points. This technique has been widely used to monitor seismic velocities in the Earth and succeeded in detecting changes in association with large earthquakes and/or volcanic eruptions. However, in doing such monitoring, it is important to judge if a current seismic velocity change is significant or not. We here propose a statistical method for that purpose. First, we study statistical distributions of seismic velocity changes observed during normal periods when no large earthquakes or volcanic eruptions are known to have occurred. Then, we assign a probability to a current value of seismic velocity change using these statistical distributions. Accordingly, we can objectively judge if the current value is normal or abnormal. Analyzing three different data sets of seismic velocities measured in Japan with seismic interferometry, we find that the Gaussian distribution well explains most of the datasets. However, an exception is the truncated Cauchy distribution that accounts for the dataset in lower frequency bands at Sakurajima volcano. Once the statistical distribution is known, whichever it is Gaussian or other distributions, we can quantify the monitoring of seismic velocity changes based on probabilities. That is also useful for automatic detections of anomalies in seismic velocity changes.