This paper focuses on observability, one of the open issues in High-Level test generation. Three different approximate metrics for taking observability into account during RT-level ATPG are presented. Metrics range from a really naïve and optimistic one to more sophisticated analysis. Metrics are evaluated including them into the calculation of the fitness function used in a RT-level ATPG. Advantages and disadvantages are illustrated. Experimental results show how sharp observability metrics are crucial for making effective RT-level ATPG possible: test sequences generated at RT-level outperform commercial gate-level ATPGs on some ITC99 benchmark circuits.