In regression analysis, a special and most attention must be paid on assumption validating the model obtained from a sample drawn form a population. The ultimate objective of any regression model is in fact, making some inference on the population. To use the model effectively to discover ( make inference) the unknown population, the model should have passed through many tests to make sure that it is the best model for the set of data representing. These tests are in fact the assumption on regression model or assumption in regression analysis. Among assumptions we have the property of having the variance of the response constant, and if this assumption is not met then the model will suffer from what we call heterogeneity or the variance of the response is not constant. The paper discussing the conceptual approaches to detect the violation of assumptions and the recovery of these violations without going to any calculation or application. The aim of this paper is to the importance of these assumptions and enhance the sense of good understanding of things will lead to better utility and rich information.