Over the years, the distributed database has been developed so fast that there's a need to develop an effective selection algorithm for it. Loo et. al. (2002) has proposed a statistical selection algorithm with the same objective and run in multicast / broadcast environment that has been proved that it is the best among others in terms of the number of messages needed to complete the searching process. However, this algorithm has a high probability of failure. A few improvements have been done to this original algorithm. This improved algorithm is developed based on the simulation of the real multicast environment. Modifications have been added in the improved algorithm to ensure that the unique pivot that never been used before is selected every time, and to solve problem that involve rank for certain key value that occur in more than one participant. Four performance measures have been conducted for the purpose of performance analysis between original and improved algorithm. These measures include probability of failure, number of messages needed, number of rounds needed and execution time. As a result, the probability of failure for the newly improved algorithm is 3.2% while the original algorithm is 19.2% without much overhead in increasing the number of messages and number of rounds needed.