Consider a life testing experiment in which (Xl, X2,…,Xn ) are such that (n—1) of them are distributed as f(x, σ)=(l/σ)e-x/σ, x ≥ σ, > 0 and one of them is distributed as f(x, σ/α), 0 < α< l. A priori each Xi has probability 1/n of being a spurious observation distributed as f (x,σ/α). For such an experiment Kale and Sinha [2] showed that if ur denotes the probability that X(r) , the rth component of the order statistic, corresponds to the spurious observation, then u1 < u2 <… < un .