It has been known for 30 years that pseudorandom number generators in the class of linear congruential generators (LCG) exhibit strong and predictable regularities. A widely used generator in this class is drand48. While the regularity is not problematic in most applications, I show that it can produce very misleading results in testing geometric algorithms that involve determinant computations. By presenting scenarios where LCG behave ‘nonrandom’ (sometimes in a spectacular way), I want to raise awareness for possible problems with LCG and pseudorandom numbers in general.