diff --git a/scheduler/rank.go b/scheduler/rank.go
index 70a6208ef..86106cf8a 100644
--- a/scheduler/rank.go
+++ b/scheduler/rank.go
@@ -1,6 +1,10 @@
 package scheduler
 
-import "github.com/hashicorp/nomad/nomad/structs"
+import (
+	"math"
+
+	"github.com/hashicorp/nomad/nomad/structs"
+)
 
 // Rank is used to provide a score and various ranking metadata
 // along with a node when iterating. This state can be modified as
@@ -108,3 +112,40 @@ func (iter *BinPackIterator) Next() *RankedNode {
 		return option
 	}
 }
+
+// scoreFit is used to score the fit based on the Google work published here:
+// http://www.columbia.edu/~cs2035/courses/ieor4405.S13/datacenter_scheduling.ppt
+// This is equivalent to their BestFit v3
+func scoreFit(node *structs.Node, util *structs.Resources) float64 {
+	// Determine the node availability
+	nodeCpu := node.Resources.CPU
+	if node.Reserved != nil {
+		nodeCpu -= node.Reserved.CPU
+	}
+	nodeMem := float64(node.Resources.MemoryMB)
+	if node.Reserved != nil {
+		nodeMem -= float64(node.Reserved.MemoryMB)
+	}
+
+	// Compute the free percentage
+	freePctCpu := 1 - (util.CPU / nodeCpu)
+	freePctRam := 1 - (float64(util.MemoryMB) / nodeMem)
+
+	// Total will be "maximized" the smaller the value is.
+	// At 100% utilization, the total is 2, while at 0% util it is 20.
+	total := math.Pow(10, freePctCpu) + math.Pow(10, freePctRam)
+
+	// Invert so that the "maximized" total represents a high-value
+	// score. Because the floor is 20, we simply use that as an anchor.
+	// This means at a perfect fit, we return 18 as the score.
+	score := 20.0 - total
+
+	// Bound the score, just in case
+	// If the score is over 18, that means we've overfit the node.
+	if score > 18.0 {
+		score = 18.0
+	} else if score < 0 {
+		score = 0
+	}
+	return score
+}
diff --git a/scheduler/rank_test.go b/scheduler/rank_test.go
index 16a75b992..b2c7775e3 100644
--- a/scheduler/rank_test.go
+++ b/scheduler/rank_test.go
@@ -30,3 +30,45 @@ func TestFeasibleRankIterator(t *testing.T) {
 		t.Fatalf("bad: %v", out)
 	}
 }
+
+func TestScoreFit(t *testing.T) {
+	node := mock.Node()
+	node.Resources = &structs.Resources{
+		CPU:      4096,
+		MemoryMB: 8192,
+	}
+	node.Reserved = &structs.Resources{
+		CPU:      2048,
+		MemoryMB: 4096,
+	}
+
+	// Test a perfect fit
+	util := &structs.Resources{
+		CPU:      2048,
+		MemoryMB: 4096,
+	}
+	score := scoreFit(node, util)
+	if score != 18.0 {
+		t.Fatalf("bad: %v", score)
+	}
+
+	// Test the worst fit
+	util = &structs.Resources{
+		CPU:      0,
+		MemoryMB: 0,
+	}
+	score = scoreFit(node, util)
+	if score != 0.0 {
+		t.Fatalf("bad: %v", score)
+	}
+
+	// Test a mid-case scenario
+	util = &structs.Resources{
+		CPU:      1024,
+		MemoryMB: 2048,
+	}
+	score = scoreFit(node, util)
+	if score < 10.0 || score > 16.0 {
+		t.Fatalf("bad: %v", score)
+	}
+}