Unpredictable dividend growth by the dividend-price ratio is considered a 'stylized fact' in post war US data. Using long-term data, covering more than 80 years from the US and three European countries, we revisit this stylized fact, and we also report results on return predictability. We find large cross-country differences regarding return and dividend growth predictability. For the US, we confirm Chen's (2008) finding of a 'tale of two periods' but with the important difference that short- and long-horizon real returns are significantly predictable in both sub-periods (1871- 1949 and 1950-2008), while long-horizon real dividend growth is unpredictable in the early period and significantly predictable in the 'wrong' direction in the post war period. These results are directly opposite to those reported by Chen using nominal returns and dividend growth. For the UK, the results are more or less similar to those for the US. For Sweden and Denmark we find no evidence of return predictability, but strong evidence of predictable dividend growth in the 'right' direction on both short and long horizons and over both the full sample periods and the post war period. We also document that implied long-horizon coefficients from VAR's often differ substantially from direct estimates in multi-year regres- sions. Throughout, we report both standard asymptotic tests and simulated small- sample tests and, following Cochrane (2008), we investigate the joint distribution of dividend-price ratio coefficients in return and dividend growth regressions.