Item cloning is increasingly used to generate slight differences in tasks for use in psychological experiments and educational assessments. This paper investigates the psychometric issues that arise when item cloning introduces variation into the difficulty parameters of the item clones. Four models are proposed and evaluated in simulation studies with conditions representing possible types of variation due to item cloning. Depending on the model specified, unaccounted variance in the item clone difficulties propagates to other parameters in the model, causing specific and predictable patterns of bias. Person parameters are largely unaffected by the choice of model, but for inferences related to the item parameters, the choice is critical and can even be leveraged to identify problematic item cloning.