How to decrease the observation variables for strong planning under partial observation is explored. Beginning from a domain under no observation, add necessary observation variables gradually to get a minimal set of observation variables necessary. Two methods are presented to decrease observation variables. With the former, when any of the two distinct states of the domain can be distinguished by an observation variable, this algorithm can find a minimal set of observation variables necessary for the execution of a plan. With the latter, when there are states that can’t be distinguished by only one observation variable, this algorithm can find a set of observation variables as small as possible which are necessary for the execution of a plan.