The expected present value of dividends is one of the classical stability criteria in actuarial risk theory. In this context, numerous papers considered threshold (refractive) and barrier (reflective) dividend strategies. These were shown to be optimal in a number of different contexts for bounded and unbounded payout rates, respectively.In this paper, motivated by the behaviour of some dividend paying stock exchange companies, we determine the optimal dividend strategy when both continuous (refractive) and lump sum (reflective) dividends can be paid at any time, and if they are subject to different transaction rates.We consider the general family of spectrally positive Lévy processes. Using scale functions, we obtain explicit formulas for the expected present value of dividends until ruin, with a penalty at ruin. We develop a verification lemma, and show that a two-layer (a, b) strategy is optimal. Such a strategy pays continuous dividends when the surplus exceeds level a > 0, and all of the excess over b > a as lump sum dividend payments. Results are illustrated.One has to determine when and how much dividends to pay-the answer to which is referred to as a dividend strategy. Traditionally, researchers focused on determining the dividend strategy that maximises the expected present value of dividends (without modifications). Main references are Gerber (1969); Loeffen (2008) for the Cramér-Lundberg model, and Bayraktar and Egami (2008); Avanzi, Shen, and Wong (2011); Bayraktar, Kyprianou, and Yamazaki (2013) for the dual model. In absence of fixed transaction costs, such optimal strategies are often of the barrier (more generally, band) type if dividend payouts are not constrained (see Bayraktar, Kyprianou, and Yamazaki, 2013, for the dual model), and of threshold type if dividend payouts are subject to a maximum payment rate (see Yin and Wen, 2013). These two main types lead to reflection or refraction strategies, respectively; see also Gerber and Shiu (2006). In presence of fixed transaction costs, impulse strategies appear (see, e.g. Bayraktar, Kyprianou, and Yamazaki, 2014).Often, reflection and refraction strategies are studied separately. However, a combination of both can be observed sometimes in practice, and can be justified from an economic point of view; see also Gerber and Loisel (2012), who argue that companies often have distribution strategies that have barrier and threshold components. In essence, the dividend optimisation determines whether surplus dollars are best distributed now, or kept for generating dividends in the future. The existence of a surplus (a buffer) is justified by the existence of frictional costs between capital injections and dividends. In other words, the existence of transaction costs (taxes, financial intermediaries, delays) mean that one should minimise any back and forth movements between companies and their investors. Because of this, if there is not much free money available (if the surplus is low) then it makes sense to minimise distribution. If there is a healthy cush...