The effect of advertising in a major product category (petrol) on the market share of a brand (petrol produced by SASOL of South Africa) was estimated using variants of a Koyck distributed lag model. Our use of a single-equation model was supported by Granger causality tests showing that advertising caused sales, but that sales did not simultaneously cause advertising. The situation is a unique natural experiment in that non-promotional variables in the marketing mix (price, location of sale and product characteristics) are largely invariant between SASOL's petrol and competing brands, allowing isolation of the effect of advertising on market share. Dummy variables were used to allow for the varying effect of different advertising campaigns and thus allow for the quality of advertising copy. The results supported the hypotheses that sales are determined by past as well as current advertising expenditures, but that the cumulative effects of advertising lasts for months rather than years. The major finding was that the quality of advertising was much more important than the quantity: variation in the quality of advertising copy was an order of magnitude more important than the effect of advertising expenditure.