diff --git a/pkg/gtr/builder.go b/pkg/gtr/builder.go
index c20212c..69323ad 100644
--- a/pkg/gtr/builder.go
+++ b/pkg/gtr/builder.go
@@ -7,9 +7,10 @@ import (
 
 // ReportBuilder builds Reports.
 type ReportBuilder struct {
-	packages   []Package
-	tests      map[int]Test
-	benchmarks map[int]Benchmark
+	packages    []Package
+	tests       map[int]Test
+	benchmarks  map[int]Benchmark
+	buildErrors map[int]BuildError
 
 	// state
 	nextId   int // next free id
@@ -25,6 +26,7 @@ func NewReportBuilder(packageName string) *ReportBuilder {
 	return &ReportBuilder{
 		tests:       make(map[int]Test),
 		benchmarks:  make(map[int]Benchmark),
+		buildErrors: make(map[int]BuildError),
 		nextId:      1,
 		packageName: packageName,
 	}
@@ -76,7 +78,29 @@ func (b *ReportBuilder) Benchmark(name string, iterations int64, nsPerOp, mbPerS
 	}
 }
 
-func (b *ReportBuilder) CreatePackage(name string, duration time.Duration) {
+func (b *ReportBuilder) CreateBuildError(packageName string) {
+	b.buildErrors[b.newId()] = BuildError{Name: packageName}
+}
+
+func (b *ReportBuilder) CreatePackage(name string, duration time.Duration, data string) {
+	// Build errors are treated somewhat differently. Rather than having a
+	// single package with all build errors collected so far, we only care
+	// about the build errors for this particular package.
+	for id, buildErr := range b.buildErrors {
+		if buildErr.Name == name {
+			if len(b.tests) > 0 || len(b.benchmarks) > 0 {
+				panic("unexpected tests and/or benchmarks found in build error package")
+			}
+			buildErr.Cause = data
+			b.packages = append(b.packages, Package{
+				Name:       name,
+				BuildError: buildErr,
+			})
+			delete(b.buildErrors, id)
+			return
+		}
+	}
+
 	// Collect tests and benchmarks for this package, maintaining insertion order.
 	var tests []Test
 	var benchmarks []Benchmark
@@ -115,9 +139,20 @@ func (b *ReportBuilder) AppendOutput(line string) {
 		b.output = append(b.output, line)
 		return
 	}
-	t := b.tests[b.lastId]
-	t.Output = append(t.Output, line)
-	b.tests[b.lastId] = t
+
+	if t, ok := b.tests[b.lastId]; ok {
+		t.Output = append(t.Output, line)
+		b.tests[b.lastId] = t
+	} else if bm, ok := b.benchmarks[b.lastId]; ok {
+		bm.Output = append(bm.Output, line)
+		b.benchmarks[b.lastId] = bm
+	} else if be, ok := b.buildErrors[b.lastId]; ok {
+		be.Output = append(be.Output, line)
+		b.buildErrors[b.lastId] = be
+	} else {
+		fmt.Printf("DEBUG output else\n")
+		b.output = append(b.output, line)
+	}
 }
 
 func (b *ReportBuilder) findTest(name string) int {
diff --git a/pkg/gtr/gtr.go b/pkg/gtr/gtr.go
index 998badc..e9e9945 100644
--- a/pkg/gtr/gtr.go
+++ b/pkg/gtr/gtr.go
@@ -38,6 +38,8 @@ type Package struct {
 
 	Tests      []Test
 	Benchmarks []Benchmark
+
+	BuildError BuildError
 }
 
 type Test struct {
@@ -58,6 +60,12 @@ type Benchmark struct {
 	AllocsPerOp int64
 }
 
+type BuildError struct {
+	Name   string
+	Cause  string
+	Output []string
+}
+
 // FromEvents creates a Report from the given list of events.
 // TODO: make packageName optional option
 func FromEvents(events []Event, packageName string) Report {
@@ -72,9 +80,11 @@ func FromEvents(events []Event, packageName string) Report {
 			report.Benchmark(ev.Name, ev.Iterations, ev.NsPerOp, ev.MBPerSec, ev.BytesPerOp, ev.AllocsPerOp)
 		case "status": // ignore for now
 		case "summary":
-			report.CreatePackage(ev.Name, ev.Duration)
+			report.CreatePackage(ev.Name, ev.Duration, ev.Data)
 		case "coverage":
 			report.Coverage(ev.CovPct, ev.CovPackages)
+		case "build_output":
+			report.CreateBuildError(ev.Name)
 		case "output":
 			report.AppendOutput(ev.Data)
 		default:
@@ -101,6 +111,21 @@ func JUnit(report Report) junit.Testsuites {
 			}
 		}
 
+		// JUnit doesn't have a good way of dealing with build errors, so we
+		// create a single failing test that contains the build error details.
+		if pkg.BuildError.Name != "" {
+			tc := junit.Testcase{
+				Classname: pkg.BuildError.Name,
+				Name:      pkg.BuildError.Cause,
+				Time:      junit.FormatDuration(0),
+				Failure: &junit.Result{
+					Message: "Failed",
+					Data:    strings.Join(pkg.BuildError.Output, "\n"),
+				},
+			}
+			suite.AddTestcase(tc)
+		}
+
 		for _, test := range pkg.Tests {
 			duration += test.Duration
 
diff --git a/pkg/parser/gotest/gotest.go b/pkg/parser/gotest/gotest.go
index b82bf91..b79a5a6 100644
--- a/pkg/parser/gotest/gotest.go
+++ b/pkg/parser/gotest/gotest.go
@@ -54,6 +54,13 @@ func (p *parser) parseLine(line string) {
 		p.coverage(matches[1], matches[2])
 	} else if matches := regexBenchmark.FindStringSubmatch(line); len(matches) == 7 {
 		p.benchmark(matches[1], matches[2], matches[3], matches[4], matches[5], matches[6])
+	} else if strings.HasPrefix(line, "# ") {
+		fields := strings.Fields(strings.TrimPrefix(line, "# "))
+		if len(fields) == 1 || len(fields) == 2 {
+			p.buildOutput(fields[0])
+		} else {
+			p.output(line)
+		}
 	} else {
 		p.output(line)
 	}
@@ -125,6 +132,13 @@ func (p *parser) benchmark(name, iterations, nsPerOp, mbPerSec, bytesPerOp, allo
 	})
 }
 
+func (p *parser) buildOutput(packageName string) {
+	p.add(gtr.Event{
+		Type: "build_output",
+		Name: packageName,
+	})
+}
+
 func (p *parser) output(line string) {
 	p.add(gtr.Event{Type: "output", Data: line})
 }