The bright type I Seyfert galaxy NGC 3516 was observed by {it Suzaku} twice, in 2005 October 12--15 and 2009 October 28--November 2, for a gross time coverage of 242 and 544 ksec and a net exposure of 134 and 255 ksec, respectively. The 2--10 keV luminosity was $2.8 times 10^{41}$ erg s$^{-1}$ in 2005, and $1.6 times 10^{41}$ erg s$^{-1}$ in 2009. The 1.4--1.7 keV and 2--10 keV count rates both exhibited peak-to-peak variations by a factor of $sim2$ in 2005, while $sim 4$ in 2009. In either observation, the 15--45 keV count rate was less variable. The 2--10 keV spectrum in 2005 was significantly more convex than that in 2009. Through a count-count-plot technique, the 2--45 keV signals in both data were successfully decomposed in a model-independent way into two distinct broadband components. One is a variable emission with a featureless spectral shape, and the other is a non-varying hard component accompanied by a prominent Fe-K emission line at 6.33 keV (6.40 keV in the rest frame). The former was fitted successfully by an absorbed power-law model, while the latter requires a new hard continuum in addition to a reflection component from distant materials. The spectral and variability differences between the two observations are mainly attributed to long-term changes of this new hard continuum, which was stable on time scales of several hundreds ksec.