Cluster analysis is fairly complex, and there are no exact guidelines for cutting the dendrogram. It is a combination of looking for large changes in the difference/similarity numbers as you change the number of clusters. I recommend plotting the number of clusters on the x-axis versus the similarity or difference values on the y-axis. When you see a sudden shift in the plot, you have a possible clustering. You then look to see whether this make logical sense.
In your graph, you start with 6 clusters. If you slice the dendrogram horizontally, there is one natural set of 3 clusters, and another at 2 clusters. The 2 cluster set is the strongest as noted by the larger vertical distance. You must then interpret the meaning of these clusters, and decide which level works best.
The pics below are an example where the same data set was used to cut the dendrogram at two different levels. Both make sense, so the option depends on your needs.