A　maximum　algebra　of　random　variables,　which　can　be　used　to　analyse　the　manufacturing　processes　in　the　discrete　event　stochastic　dynamic　systems,　is　presented.　A　kind　of　series-parallel　production　line　systems　in　which　the　work　durations　of　all　tasks　are　random　variables　with　exponential　distributions　is　discussed.　By　means　of　the　maximum　algebra　of　random　variables　the　statistical　characters　of　total　manufacturing　time　are　abtained　in　order　to　estimate　its　production　performance.　An　example　is　given　for　confirming　the　validity　of　the　method　by　camparing　the　result　of　simulation.